script-2-Ukraine-tweets.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "provenance": [],
      "mount_file_id": "1ogY7HZ71hX54eWn8651HIZkjlI6c2YcC",
      "authorship_tag": "ABX9TyPvcCgd55PkCsyqcPuoYuWE",
      "include_colab_link": true
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    },
    "language_info": {
      "name": "python"
    },
    "accelerator": "GPU",
    "widgets": {
      "application/vnd.jupyter.widget-state+json": {
        "49d3cbe48b8a4d389c328ffb28268d34": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "HBoxModel",
          "model_module_version": "1.5.0",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "HBoxModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "HBoxView",
            "box_style": "",
            "children": [
              "IPY_MODEL_93c7a3c727d94661a55b16da5357ef01",
              "IPY_MODEL_83bb56d48a1b43208db866d4766cd5d7",
              "IPY_MODEL_09256a2b562d4caa93df6245f6969abf"
            ],
            "layout": "IPY_MODEL_fe48db3d6f4140c4a5e2d8a2895cce43"
          }
        },
        "93c7a3c727d94661a55b16da5357ef01": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "HTMLModel",
          "model_module_version": "1.5.0",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "HTMLModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "HTMLView",
            "description": "",
            "description_tooltip": null,
            "layout": "IPY_MODEL_c69882ee465d477cbca0f1acca88d68d",
            "placeholder": "​",
            "style": "IPY_MODEL_bf07759821524a808553b7834d68a989",
            "value": "Downloading: 100%"
          }
        },
        "83bb56d48a1b43208db866d4766cd5d7": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "FloatProgressModel",
          "model_module_version": "1.5.0",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "FloatProgressModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "ProgressView",
            "bar_style": "success",
            "description": "",
            "description_tooltip": null,
            "layout": "IPY_MODEL_da0aecff3b3c4784b9a089bd01d08802",
            "max": 231508,
            "min": 0,
            "orientation": "horizontal",
            "style": "IPY_MODEL_a7e7c9860f484990af5673058da0148a",
            "value": 231508
          }
        },
        "09256a2b562d4caa93df6245f6969abf": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "HTMLModel",
          "model_module_version": "1.5.0",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "HTMLModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "HTMLView",
            "description": "",
            "description_tooltip": null,
            "layout": "IPY_MODEL_76c2869f64924ae6b19f970e601a0d73",
            "placeholder": "​",
            "style": "IPY_MODEL_814cfe825aa94b67aca149171c2a2b1d",
            "value": " 232k/232k [00:00&lt;00:00, 244kB/s]"
          }
        },
        "fe48db3d6f4140c4a5e2d8a2895cce43": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "model_module_version": "1.2.0",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": null,
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "c69882ee465d477cbca0f1acca88d68d": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "model_module_version": "1.2.0",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": null,
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "bf07759821524a808553b7834d68a989": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "DescriptionStyleModel",
          "model_module_version": "1.5.0",
          "state": {
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "DescriptionStyleModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "StyleView",
            "description_width": ""
          }
        },
        "da0aecff3b3c4784b9a089bd01d08802": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "model_module_version": "1.2.0",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": null,
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "a7e7c9860f484990af5673058da0148a": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "ProgressStyleModel",
          "model_module_version": "1.5.0",
          "state": {
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "ProgressStyleModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "StyleView",
            "bar_color": null,
            "description_width": ""
          }
        },
        "76c2869f64924ae6b19f970e601a0d73": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "model_module_version": "1.2.0",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": null,
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "814cfe825aa94b67aca149171c2a2b1d": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "DescriptionStyleModel",
          "model_module_version": "1.5.0",
          "state": {
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "DescriptionStyleModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "StyleView",
            "description_width": ""
          }
        },
        "5a10256ef3ee44839e37f84a8400b7e5": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "HBoxModel",
          "model_module_version": "1.5.0",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "HBoxModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "HBoxView",
            "box_style": "",
            "children": [
              "IPY_MODEL_cbf63039110d4a7181900cea3fe02e81",
              "IPY_MODEL_ccc9f853d1d74a9e9048647491ea78a3",
              "IPY_MODEL_1c093f2473154b9aa219fa7617f96e52"
            ],
            "layout": "IPY_MODEL_41ff7cbe96a341dcafba8d60902be9b1"
          }
        },
        "cbf63039110d4a7181900cea3fe02e81": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "HTMLModel",
          "model_module_version": "1.5.0",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "HTMLModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "HTMLView",
            "description": "",
            "description_tooltip": null,
            "layout": "IPY_MODEL_cd0f776c715646c3ab1aba589834a0c5",
            "placeholder": "​",
            "style": "IPY_MODEL_40b260b7c75240efa8c73ab72cf51742",
            "value": "Downloading: 100%"
          }
        },
        "ccc9f853d1d74a9e9048647491ea78a3": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "FloatProgressModel",
          "model_module_version": "1.5.0",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "FloatProgressModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "ProgressView",
            "bar_style": "success",
            "description": "",
            "description_tooltip": null,
            "layout": "IPY_MODEL_ef579abff8d44263811bc4758412abfe",
            "max": 28,
            "min": 0,
            "orientation": "horizontal",
            "style": "IPY_MODEL_8d7966ee53544cfb8daed844880bf74e",
            "value": 28
          }
        },
        "1c093f2473154b9aa219fa7617f96e52": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "HTMLModel",
          "model_module_version": "1.5.0",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "HTMLModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "HTMLView",
            "description": "",
            "description_tooltip": null,
            "layout": "IPY_MODEL_161c3aaa29324c558a485531cc19b5ba",
            "placeholder": "​",
            "style": "IPY_MODEL_c78ab646d95f49d88ceff2a484e4aa1e",
            "value": " 28.0/28.0 [00:00&lt;00:00, 429B/s]"
          }
        },
        "41ff7cbe96a341dcafba8d60902be9b1": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "model_module_version": "1.2.0",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": null,
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "cd0f776c715646c3ab1aba589834a0c5": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "model_module_version": "1.2.0",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": null,
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "40b260b7c75240efa8c73ab72cf51742": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "DescriptionStyleModel",
          "model_module_version": "1.5.0",
          "state": {
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "DescriptionStyleModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "StyleView",
            "description_width": ""
          }
        },
        "ef579abff8d44263811bc4758412abfe": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "model_module_version": "1.2.0",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": null,
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "8d7966ee53544cfb8daed844880bf74e": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "ProgressStyleModel",
          "model_module_version": "1.5.0",
          "state": {
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "ProgressStyleModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "StyleView",
            "bar_color": null,
            "description_width": ""
          }
        },
        "161c3aaa29324c558a485531cc19b5ba": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "model_module_version": "1.2.0",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": null,
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "c78ab646d95f49d88ceff2a484e4aa1e": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "DescriptionStyleModel",
          "model_module_version": "1.5.0",
          "state": {
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "DescriptionStyleModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "StyleView",
            "description_width": ""
          }
        },
        "94c8d74ea72541218695bd3e8e1cb533": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "HBoxModel",
          "model_module_version": "1.5.0",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "HBoxModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "HBoxView",
            "box_style": "",
            "children": [
              "IPY_MODEL_b9257345b568489e8d4a4e4a6327d85e",
              "IPY_MODEL_ff2a7cb1b7b14bc4ab2531947a4d9bd1",
              "IPY_MODEL_73ec91fb2cd24e86b481319881fb18ca"
            ],
            "layout": "IPY_MODEL_05745987b5e74b76b986a42fff94d988"
          }
        },
        "b9257345b568489e8d4a4e4a6327d85e": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "HTMLModel",
          "model_module_version": "1.5.0",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "HTMLModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "HTMLView",
            "description": "",
            "description_tooltip": null,
            "layout": "IPY_MODEL_295e095abd2342658f3d7831828c0050",
            "placeholder": "​",
            "style": "IPY_MODEL_b2b5e03887c244d9abcca2651df49615",
            "value": "Downloading: 100%"
          }
        },
        "ff2a7cb1b7b14bc4ab2531947a4d9bd1": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "FloatProgressModel",
          "model_module_version": "1.5.0",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "FloatProgressModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "ProgressView",
            "bar_style": "success",
            "description": "",
            "description_tooltip": null,
            "layout": "IPY_MODEL_dd7ba37072b3442a9262e6028820f4c9",
            "max": 570,
            "min": 0,
            "orientation": "horizontal",
            "style": "IPY_MODEL_83c6bc25306b4b9dbd5fa789c2eaeeda",
            "value": 570
          }
        },
        "73ec91fb2cd24e86b481319881fb18ca": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "HTMLModel",
          "model_module_version": "1.5.0",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "HTMLModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "HTMLView",
            "description": "",
            "description_tooltip": null,
            "layout": "IPY_MODEL_0f9c308c59824e11844db61b37773df7",
            "placeholder": "​",
            "style": "IPY_MODEL_963b6d1d1e794ecfb090ea34a572fdf5",
            "value": " 570/570 [00:00&lt;00:00, 6.26kB/s]"
          }
        },
        "05745987b5e74b76b986a42fff94d988": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "model_module_version": "1.2.0",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": null,
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "295e095abd2342658f3d7831828c0050": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "model_module_version": "1.2.0",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": null,
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "b2b5e03887c244d9abcca2651df49615": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "DescriptionStyleModel",
          "model_module_version": "1.5.0",
          "state": {
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "DescriptionStyleModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "StyleView",
            "description_width": ""
          }
        },
        "dd7ba37072b3442a9262e6028820f4c9": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "model_module_version": "1.2.0",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": null,
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "83c6bc25306b4b9dbd5fa789c2eaeeda": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "ProgressStyleModel",
          "model_module_version": "1.5.0",
          "state": {
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "ProgressStyleModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "StyleView",
            "bar_color": null,
            "description_width": ""
          }
        },
        "0f9c308c59824e11844db61b37773df7": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "model_module_version": "1.2.0",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": null,
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "963b6d1d1e794ecfb090ea34a572fdf5": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "DescriptionStyleModel",
          "model_module_version": "1.5.0",
          "state": {
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "DescriptionStyleModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "StyleView",
            "description_width": ""
          }
        }
      }
    }
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "view-in-github",
        "colab_type": "text"
      },
      "source": [
        "<a href=\"https://colab.research.google.com/github/fmigas/Projects/blob/main/Ukraine_tweets.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "There are are couple of reasons why I decided to set up my own project rather than use a public dataset or join a kaggle competition:\n",
        "\n",
        "\n",
        "*   I decided to define a problem myself\n",
        "*   I needed to collect data myself\n",
        "*   I could not follow any other kaggle notebooks to find the best solution ;-)\n",
        "*   I needed to make my own decisions on all key factors, like project goal and metrics\n",
        "*   I had no clear benchmark to find out \"what is a good result\"\n",
        "\n",
        "In other words, it looks more like a \"real world\" project than an artificial kaggle competition, whare data, objectives, metrics and benchmarks are defined.\n",
        "\n"
      ],
      "metadata": {
        "id": "ziP0-rj_CTyp"
      }
    },
    {
      "cell_type": "markdown",
      "source": [
        "**Problem definition:** take a collection of tweets on Ukraine and find out, which of them come from an \"expert\".\n",
        "\n",
        "The measure of \"being an expert\" is here a number of followers. I'll discuss further what threshold (a thousand of followers? ten thousand?) I decided to pick.\n",
        "\n",
        "I am aware this problem may seem trivial and a bit unrealistic, but the big advantage is that I did not have to care about labelled data. When parsing tweets, we can easily get a number of the user's followers."
      ],
      "metadata": {
        "id": "P_XuxfaFEpmZ"
      }
    },
    {
      "cell_type": "markdown",
      "source": [
        "**First step - data collection**\n",
        "\n",
        "I used snscrape module to download 100.000 tweets containing a word \"Ukraine\", all written in English.\n",
        "I downloaded 4.000 tweets each day from 2nd of October through 27 of October."
      ],
      "metadata": {
        "id": "HZI17ze2JC8y"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "import snscrape.modules.twitter as sntwitter\n",
        "import pandas as pd\n",
        "\n",
        "tweets_list = []\n",
        "for n in range(2, 27):\n",
        "    date = '2022-09-' + str(n)\n",
        "    query = 'Ukraine lang:en until:' + date\n",
        "    print(query)\n",
        "    # Using TwitterSearchScraper to scrape data and append tweets to list\n",
        "    for i, tweet in enumerate(sntwitter.TwitterSearchScraper(query).get_items()):\n",
        "        if i >= 4000:\n",
        "            break\n",
        "        tweets_list.append([tweet.date, tweet.content, tweet.user.username, tweet.user.followersCount])\n",
        "\n",
        "# Creating a dataframe from the tweets list above\n",
        "df = pd.DataFrame(tweets_list, columns = ['Date', 'Text', 'Username', 'followersCount'])"
      ],
      "metadata": {
        "id": "Dfsn4HfqCZyA"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "source": [
        "**Raw ideas**\n",
        "\n",
        "Before EDA and data processing, let's ask a couple of questions and extract a couple of features from the raw text.\n",
        "\n",
        "What text features may make difference between an \"expert opinion/message\" and a \"non-expert comment\"?\n",
        "* Maybe \"expert\" messages are longer?\n",
        "* Or maybe they are use slightly different vocabulary?\n",
        "* Or maybe \"non-experts\" abuse dirty words in their comments?\n",
        "* Or maybe \"experts\" include more hashes or links in their messages?\n",
        "* Or should we dig deeper into Natural Language Understanding tools, like contextual embeddings from the state-of-the-art transformer models like BERT?\n",
        "\n",
        "\n",
        "\n"
      ],
      "metadata": {
        "id": "nxsHQA8p1Rhu"
      }
    },
    {
      "cell_type": "markdown",
      "source": [
        "**\"Expert\" definition**\n",
        "My initial notion was to define an \"expert\" as a Twitter use with at least 10.000 followers.\n",
        "\n",
        "At the end, we'll check alternative thresholds (from 1.000 to 10.000, with a 1.000 step) and we'll observe how a modified definition affects the results."
      ],
      "metadata": {
        "id": "gZL7mosF9Jgf"
      }
    },
    {
      "cell_type": "markdown",
      "source": [
        "**Feature extraction**\n",
        "Before we clean and tokenize tweets, let's extract a couple of features:\n",
        "* tweet length\n",
        "* '#' count in a tweet\n",
        "* @ count in a tweet\n",
        "* dirty words count in a tweet\n",
        "* links count in a tweet"
      ],
      "metadata": {
        "id": "GVvJhGwP-biX"
      }
    },
    {
      "cell_type": "markdown",
      "source": [
        "Let's start with loading a raw file from the repository."
      ],
      "metadata": {
        "id": "b8mfSr3SAPKY"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "import pandas as pd\n",
        "import numpy as np\n",
        "\n",
        "df = pd.read_csv(\"/content/drive/MyDrive/Python_data/tweets.csv\")\n",
        "df.drop(['Unnamed: 0'], inplace = True, axis = 1)\n",
        "df.isna().sum()\n",
        "df.drop_duplicates(inplace = True)\n",
        "df.dropna(inplace = True)\n",
        "df.columns = 'date text user num'.split()\n",
        "df.num = df.num.astype('int32')\n",
        "\n",
        "\n",
        "df['at'] = df['text'].apply(lambda x: x.count('@'))\n",
        "df['hash'] = df['text'].apply(lambda x: x.count('#'))\n",
        "df['len'] = df['text'].apply(lambda x: len(x))\n",
        "df['links'] = df['text'].apply(lambda x: x.count('http'))\n",
        "df['expert'] = df['num'].apply(lambda x: 1 if x > 10000 else 0)"
      ],
      "metadata": {
        "id": "4OHIhobnHiDc"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "source": [
        "We added 'at', 'hash', 'len', 'links' columns and an 'expert' column mapped intuitively: expert: 1, non-expert: 0.\n"
      ],
      "metadata": {
        "id": "JqrlqKz7BFjh"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "df.head()"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 204
        },
        "id": "tWOBbP4PBCpY",
        "outputId": "15d74c88-5efe-4dae-e81b-9723b6f37e91"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "                        date  \\\n",
              "0  2022-09-01 23:59:54+00:00   \n",
              "1  2022-09-01 23:59:47+00:00   \n",
              "2  2022-09-01 23:59:47+00:00   \n",
              "3  2022-09-01 23:59:46+00:00   \n",
              "4  2022-09-01 23:59:46+00:00   \n",
              "\n",
              "                                                text           user   num  at  \\\n",
              "0  @SenatorWicker @CNN @jimsciutto Please send 20...   ArtemiyMoroz   105   3   \n",
              "1  @Der_Parrot If anything, I see vast potential ...   AdrianEMoser     0   1   \n",
              "2  the whos feeding the sheep, this probably goes...       werchenz   107   0   \n",
              "3  look you too can support Nazis in Ukraine\\njus...     vandman777  2968   0   \n",
              "4  Create a false pandemic instilling fear &amp; ...  thelastpers0n    37   0   \n",
              "\n",
              "   hash  len  links  expert  \n",
              "0     1   80      0       0  \n",
              "1     0  287      0       0  \n",
              "2     0  270      0       0  \n",
              "3     0  110      1       0  \n",
              "4     0  307      1       0  "
            ],
            "text/html": [
              "\n",
              "  <div id=\"df-f06c2221-7306-4887-beca-6346d03766e0\">\n",
              "    <div class=\"colab-df-container\">\n",
              "      <div>\n",
              "<style scoped>\n",
              "    .dataframe tbody tr th:only-of-type {\n",
              "        vertical-align: middle;\n",
              "    }\n",
              "\n",
              "    .dataframe tbody tr th {\n",
              "        vertical-align: top;\n",
              "    }\n",
              "\n",
              "    .dataframe thead th {\n",
              "        text-align: right;\n",
              "    }\n",
              "</style>\n",
              "<table border=\"1\" class=\"dataframe\">\n",
              "  <thead>\n",
              "    <tr style=\"text-align: right;\">\n",
              "      <th></th>\n",
              "      <th>date</th>\n",
              "      <th>text</th>\n",
              "      <th>user</th>\n",
              "      <th>num</th>\n",
              "      <th>at</th>\n",
              "      <th>hash</th>\n",
              "      <th>len</th>\n",
              "      <th>links</th>\n",
              "      <th>expert</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>0</th>\n",
              "      <td>2022-09-01 23:59:54+00:00</td>\n",
              "      <td>@SenatorWicker @CNN @jimsciutto Please send 20...</td>\n",
              "      <td>ArtemiyMoroz</td>\n",
              "      <td>105</td>\n",
              "      <td>3</td>\n",
              "      <td>1</td>\n",
              "      <td>80</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1</th>\n",
              "      <td>2022-09-01 23:59:47+00:00</td>\n",
              "      <td>@Der_Parrot If anything, I see vast potential ...</td>\n",
              "      <td>AdrianEMoser</td>\n",
              "      <td>0</td>\n",
              "      <td>1</td>\n",
              "      <td>0</td>\n",
              "      <td>287</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>2</th>\n",
              "      <td>2022-09-01 23:59:47+00:00</td>\n",
              "      <td>the whos feeding the sheep, this probably goes...</td>\n",
              "      <td>werchenz</td>\n",
              "      <td>107</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>270</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>3</th>\n",
              "      <td>2022-09-01 23:59:46+00:00</td>\n",
              "      <td>look you too can support Nazis in Ukraine\\njus...</td>\n",
              "      <td>vandman777</td>\n",
              "      <td>2968</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>110</td>\n",
              "      <td>1</td>\n",
              "      <td>0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>4</th>\n",
              "      <td>2022-09-01 23:59:46+00:00</td>\n",
              "      <td>Create a false pandemic instilling fear &amp;amp; ...</td>\n",
              "      <td>thelastpers0n</td>\n",
              "      <td>37</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>307</td>\n",
              "      <td>1</td>\n",
              "      <td>0</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "</div>\n",
              "      <button class=\"colab-df-convert\" onclick=\"convertToInteractive('df-f06c2221-7306-4887-beca-6346d03766e0')\"\n",
              "              title=\"Convert this dataframe to an interactive table.\"\n",
              "              style=\"display:none;\">\n",
              "        \n",
              "  <svg xmlns=\"http://www.w3.org/2000/svg\" height=\"24px\"viewBox=\"0 0 24 24\"\n",
              "       width=\"24px\">\n",
              "    <path d=\"M0 0h24v24H0V0z\" fill=\"none\"/>\n",
              "    <path d=\"M18.56 5.44l.94 2.06.94-2.06 2.06-.94-2.06-.94-.94-2.06-.94 2.06-2.06.94zm-11 1L8.5 8.5l.94-2.06 2.06-.94-2.06-.94L8.5 2.5l-.94 2.06-2.06.94zm10 10l.94 2.06.94-2.06 2.06-.94-2.06-.94-.94-2.06-.94 2.06-2.06.94z\"/><path d=\"M17.41 7.96l-1.37-1.37c-.4-.4-.92-.59-1.43-.59-.52 0-1.04.2-1.43.59L10.3 9.45l-7.72 7.72c-.78.78-.78 2.05 0 2.83L4 21.41c.39.39.9.59 1.41.59.51 0 1.02-.2 1.41-.59l7.78-7.78 2.81-2.81c.8-.78.8-2.07 0-2.86zM5.41 20L4 18.59l7.72-7.72 1.47 1.35L5.41 20z\"/>\n",
              "  </svg>\n",
              "      </button>\n",
              "      \n",
              "  <style>\n",
              "    .colab-df-container {\n",
              "      display:flex;\n",
              "      flex-wrap:wrap;\n",
              "      gap: 12px;\n",
              "    }\n",
              "\n",
              "    .colab-df-convert {\n",
              "      background-color: #E8F0FE;\n",
              "      border: none;\n",
              "      border-radius: 50%;\n",
              "      cursor: pointer;\n",
              "      display: none;\n",
              "      fill: #1967D2;\n",
              "      height: 32px;\n",
              "      padding: 0 0 0 0;\n",
              "      width: 32px;\n",
              "    }\n",
              "\n",
              "    .colab-df-convert:hover {\n",
              "      background-color: #E2EBFA;\n",
              "      box-shadow: 0px 1px 2px rgba(60, 64, 67, 0.3), 0px 1px 3px 1px rgba(60, 64, 67, 0.15);\n",
              "      fill: #174EA6;\n",
              "    }\n",
              "\n",
              "    [theme=dark] .colab-df-convert {\n",
              "      background-color: #3B4455;\n",
              "      fill: #D2E3FC;\n",
              "    }\n",
              "\n",
              "    [theme=dark] .colab-df-convert:hover {\n",
              "      background-color: #434B5C;\n",
              "      box-shadow: 0px 1px 3px 1px rgba(0, 0, 0, 0.15);\n",
              "      filter: drop-shadow(0px 1px 2px rgba(0, 0, 0, 0.3));\n",
              "      fill: #FFFFFF;\n",
              "    }\n",
              "  </style>\n",
              "\n",
              "      <script>\n",
              "        const buttonEl =\n",
              "          document.querySelector('#df-f06c2221-7306-4887-beca-6346d03766e0 button.colab-df-convert');\n",
              "        buttonEl.style.display =\n",
              "          google.colab.kernel.accessAllowed ? 'block' : 'none';\n",
              "\n",
              "        async function convertToInteractive(key) {\n",
              "          const element = document.querySelector('#df-f06c2221-7306-4887-beca-6346d03766e0');\n",
              "          const dataTable =\n",
              "            await google.colab.kernel.invokeFunction('convertToInteractive',\n",
              "                                                     [key], {});\n",
              "          if (!dataTable) return;\n",
              "\n",
              "          const docLinkHtml = 'Like what you see? Visit the ' +\n",
              "            '<a target=\"_blank\" href=https://colab.research.google.com/notebooks/data_table.ipynb>data table notebook</a>'\n",
              "            + ' to learn more about interactive tables.';\n",
              "          element.innerHTML = '';\n",
              "          dataTable['output_type'] = 'display_data';\n",
              "          await google.colab.output.renderOutput(dataTable, element);\n",
              "          const docLink = document.createElement('div');\n",
              "          docLink.innerHTML = docLinkHtml;\n",
              "          element.appendChild(docLink);\n",
              "        }\n",
              "      </script>\n",
              "    </div>\n",
              "  </div>\n",
              "  "
            ]
          },
          "metadata": {},
          "execution_count": 3
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "One of the initial ideas was that \"non-experts\" may have a tendency to abuse dirty words in their tweets. I searched for \"English bad words\" on the internet and downloaded it. I will not print it here (I am not an English native speaker, English bad words do not resonate in me so much, but I guess it could be unpleasant for a native speaker to see such a list here).\n",
        "\n",
        "I made one modification to the badworeds list. I removed a word \"nazi\" and its variations from the list. As you probably know, Russian propaganda calls a legitimate Ukrainian government \"nazis\", and it is ofted cited by official sources. Considering it a \"dirty word\" could lead to wrong conclusions.\n",
        "\n",
        "We'll add a \"badwords count\" column later, after tweets cleaning and processing."
      ],
      "metadata": {
        "id": "Piy8dW5UBfec"
      }
    },
    {
      "cell_type": "markdown",
      "source": [
        "**Initial EDA**\n",
        "Let's start with seeing if our dataset is a balanced one."
      ],
      "metadata": {
        "id": "BxF19_6pD3I1"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "print(f'Number of records: {df.shape[0]}')\n",
        "print('Value counts - non_expert vs expert')\n",
        "df['expert'].value_counts()"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "PMOArlqZEP5D",
        "outputId": "f4130ce3-ea97-4537-9f6f-9a1bb93d8fac"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Number of records: 99992\n",
            "Value counts - non_expert vs expert\n"
          ]
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "0    93165\n",
              "1     6827\n",
              "Name: expert, dtype: int64"
            ]
          },
          "metadata": {},
          "execution_count": 4
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "It was easy to foresee - our dataset is highly skewed towards \"non expert\" values. Which is natural - in each field there are fewer experts, otherwise they would not be called \"experts\", would they?\n",
        "\n",
        "Let's take a look at a histogram of a followersCount column.\n",
        "\n",
        "However, as some users have a very large number of followers (e.g. tens of millions), any plot (histogram, boxplot) trying to cover the full span of followersCount count, would look not so much informative with one high peak for all \"low values\" bin. On the other hand, applying a log scale would make it misleading.\n",
        "\n",
        "That is why we plotted a histogram only for non-expert users, with fewer than 10.000 followers. As we can see, even in that group, the majority of users have fewer than 1.000 followers. We'll come back to the question \"how an expert should be defined\" later.\n",
        "\n",
        "*For my convenience, I renamed followersCount column to 'num', to make it shorter*"
      ],
      "metadata": {
        "id": "mCpiDQXmFCO4"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "import matplotlib.pyplot as plt\n",
        "\n",
        "plt.hist(df[df['num'] < 10000]['num'], bins = 20)\n",
        "plt.show()"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 265
        },
        "id": "Sct23v9IFobv",
        "outputId": "9c14ca0f-58af-475a-b0bd-71303d867b41"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ],
            "image/png": "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\n"
          },
          "metadata": {
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "**Text cleaning and TOKENIZATION**\n",
        "Before we follow with the remaining EDA, let's add badwords count to the pack. To do so, let's do some text cleaning. We'll use regular expression and BeautifulSoup library to proceed."
      ],
      "metadata": {
        "id": "oUKezKd_KrGb"
      }
    },
    {
      "cell_type": "markdown",
      "source": [
        "At this step, we will also do the tokenization. Out of many different tokenizers available, a natural choice for any Twitter-oriented project seems to be an NLTK TweetTokenizer.\n",
        "One of the text cleaning procedure steps will be replacing negations like \"isn't\" with their base form \"is not\" etc. We'll see later, if using bigrams that should catch such two-word phrases would add to the model quality."
      ],
      "metadata": {
        "id": "dYyZGVY6MMoh"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "import re\n",
        "from bs4 import BeautifulSoup\n",
        "from nltk.tokenize import TweetTokenizer\n",
        "tokenizer = TweetTokenizer()\n",
        "\n",
        "pat1 = r'@[A-Za-z0-9_]+'\n",
        "pat2 = r'https?://[^ ]+'\n",
        "combined_pat = r'|'.join((pat1, pat2))\n",
        "www_pat = r'www.[^ ]+'\n",
        "negations_dic = {\"isn't\":\"is not\", \"aren't\":\"are not\", \"wasn't\":\"was not\", \"weren't\":\"were not\",\n",
        "                \"haven't\":\"have not\",\"hasn't\":\"has not\",\"hadn't\":\"had not\",\"won't\":\"will not\",\n",
        "                \"wouldn't\":\"would not\", \"don't\":\"do not\", \"doesn't\":\"does not\",\"didn't\":\"did not\",\n",
        "                \"can't\":\"can not\",\"couldn't\":\"could not\",\"shouldn't\":\"should not\",\"mightn't\":\"might not\",\n",
        "                \"mustn't\":\"must not\"}\n",
        "neg_pattern = re.compile(r'\\b(' + '|'.join(negations_dic.keys()) + r')\\b')\n",
        "\n",
        "def tweet_cleaner_updated(text):\n",
        "    soup = BeautifulSoup(text, 'lxml')\n",
        "    souped = soup.get_text()\n",
        "    try:\n",
        "        bom_removed = souped.decode(\"utf-8-sig\").replace(u\"\\ufffd\", \"?\")\n",
        "    except:\n",
        "        bom_removed = souped\n",
        "    stripped = re.sub(combined_pat, '', bom_removed)\n",
        "    stripped = re.sub(www_pat, '', stripped)\n",
        "    lower_case = stripped.lower()\n",
        "    neg_handled = neg_pattern.sub(lambda x: negations_dic[x.group()], lower_case)\n",
        "    letters_only = re.sub(\"[^a-zA-Z]\", \" \", neg_handled)\n",
        "\n",
        "    words = tokenizer.tokenize(letters_only)\n",
        "    return (\" \".join(words)).strip() # we return a string, not tokens, as most models (tfidf, BERT) require str values as an input\n",
        "\n",
        "df['processed'] = df['text'].apply(lambda x: tweet_cleaner_updated(x)) "
      ],
      "metadata": {
        "id": "9sJjXSWPLZAn"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "df.head()"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 204
        },
        "id": "rVCpcESRNQv3",
        "outputId": "ed832dc6-6605-4c04-9b28-c19a51d1b6b5"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "                        date  \\\n",
              "0  2022-09-01 23:59:54+00:00   \n",
              "1  2022-09-01 23:59:47+00:00   \n",
              "2  2022-09-01 23:59:47+00:00   \n",
              "3  2022-09-01 23:59:46+00:00   \n",
              "4  2022-09-01 23:59:46+00:00   \n",
              "\n",
              "                                                text           user   num  at  \\\n",
              "0  @SenatorWicker @CNN @jimsciutto Please send 20...   ArtemiyMoroz   105   3   \n",
              "1  @Der_Parrot If anything, I see vast potential ...   AdrianEMoser     0   1   \n",
              "2  the whos feeding the sheep, this probably goes...       werchenz   107   0   \n",
              "3  look you too can support Nazis in Ukraine\\njus...     vandman777  2968   0   \n",
              "4  Create a false pandemic instilling fear &amp; ...  thelastpers0n    37   0   \n",
              "\n",
              "   hash  len  links  expert                                          processed  \n",
              "0     1   80      0       0        please send more himars to ukraine urgently  \n",
              "1     0  287      0       0  if anything i see vast potential in creating j...  \n",
              "2     0  270      0       0  the whos feeding the sheep this probably goes ...  \n",
              "3     0  110      1       0  look you too can support nazis in ukraine just...  \n",
              "4     0  307      1       0  create a false pandemic instilling fear confor...  "
            ],
            "text/html": [
              "\n",
              "  <div id=\"df-f933e963-3003-47d9-a977-a891d58130a7\">\n",
              "    <div class=\"colab-df-container\">\n",
              "      <div>\n",
              "<style scoped>\n",
              "    .dataframe tbody tr th:only-of-type {\n",
              "        vertical-align: middle;\n",
              "    }\n",
              "\n",
              "    .dataframe tbody tr th {\n",
              "        vertical-align: top;\n",
              "    }\n",
              "\n",
              "    .dataframe thead th {\n",
              "        text-align: right;\n",
              "    }\n",
              "</style>\n",
              "<table border=\"1\" class=\"dataframe\">\n",
              "  <thead>\n",
              "    <tr style=\"text-align: right;\">\n",
              "      <th></th>\n",
              "      <th>date</th>\n",
              "      <th>text</th>\n",
              "      <th>user</th>\n",
              "      <th>num</th>\n",
              "      <th>at</th>\n",
              "      <th>hash</th>\n",
              "      <th>len</th>\n",
              "      <th>links</th>\n",
              "      <th>expert</th>\n",
              "      <th>processed</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>0</th>\n",
              "      <td>2022-09-01 23:59:54+00:00</td>\n",
              "      <td>@SenatorWicker @CNN @jimsciutto Please send 20...</td>\n",
              "      <td>ArtemiyMoroz</td>\n",
              "      <td>105</td>\n",
              "      <td>3</td>\n",
              "      <td>1</td>\n",
              "      <td>80</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>please send more himars to ukraine urgently</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1</th>\n",
              "      <td>2022-09-01 23:59:47+00:00</td>\n",
              "      <td>@Der_Parrot If anything, I see vast potential ...</td>\n",
              "      <td>AdrianEMoser</td>\n",
              "      <td>0</td>\n",
              "      <td>1</td>\n",
              "      <td>0</td>\n",
              "      <td>287</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>if anything i see vast potential in creating j...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>2</th>\n",
              "      <td>2022-09-01 23:59:47+00:00</td>\n",
              "      <td>the whos feeding the sheep, this probably goes...</td>\n",
              "      <td>werchenz</td>\n",
              "      <td>107</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>270</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>the whos feeding the sheep this probably goes ...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>3</th>\n",
              "      <td>2022-09-01 23:59:46+00:00</td>\n",
              "      <td>look you too can support Nazis in Ukraine\\njus...</td>\n",
              "      <td>vandman777</td>\n",
              "      <td>2968</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>110</td>\n",
              "      <td>1</td>\n",
              "      <td>0</td>\n",
              "      <td>look you too can support nazis in ukraine just...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>4</th>\n",
              "      <td>2022-09-01 23:59:46+00:00</td>\n",
              "      <td>Create a false pandemic instilling fear &amp;amp; ...</td>\n",
              "      <td>thelastpers0n</td>\n",
              "      <td>37</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>307</td>\n",
              "      <td>1</td>\n",
              "      <td>0</td>\n",
              "      <td>create a false pandemic instilling fear confor...</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "</div>\n",
              "      <button class=\"colab-df-convert\" onclick=\"convertToInteractive('df-f933e963-3003-47d9-a977-a891d58130a7')\"\n",
              "              title=\"Convert this dataframe to an interactive table.\"\n",
              "              style=\"display:none;\">\n",
              "        \n",
              "  <svg xmlns=\"http://www.w3.org/2000/svg\" height=\"24px\"viewBox=\"0 0 24 24\"\n",
              "       width=\"24px\">\n",
              "    <path d=\"M0 0h24v24H0V0z\" fill=\"none\"/>\n",
              "    <path d=\"M18.56 5.44l.94 2.06.94-2.06 2.06-.94-2.06-.94-.94-2.06-.94 2.06-2.06.94zm-11 1L8.5 8.5l.94-2.06 2.06-.94-2.06-.94L8.5 2.5l-.94 2.06-2.06.94zm10 10l.94 2.06.94-2.06 2.06-.94-2.06-.94-.94-2.06-.94 2.06-2.06.94z\"/><path d=\"M17.41 7.96l-1.37-1.37c-.4-.4-.92-.59-1.43-.59-.52 0-1.04.2-1.43.59L10.3 9.45l-7.72 7.72c-.78.78-.78 2.05 0 2.83L4 21.41c.39.39.9.59 1.41.59.51 0 1.02-.2 1.41-.59l7.78-7.78 2.81-2.81c.8-.78.8-2.07 0-2.86zM5.41 20L4 18.59l7.72-7.72 1.47 1.35L5.41 20z\"/>\n",
              "  </svg>\n",
              "      </button>\n",
              "      \n",
              "  <style>\n",
              "    .colab-df-container {\n",
              "      display:flex;\n",
              "      flex-wrap:wrap;\n",
              "      gap: 12px;\n",
              "    }\n",
              "\n",
              "    .colab-df-convert {\n",
              "      background-color: #E8F0FE;\n",
              "      border: none;\n",
              "      border-radius: 50%;\n",
              "      cursor: pointer;\n",
              "      display: none;\n",
              "      fill: #1967D2;\n",
              "      height: 32px;\n",
              "      padding: 0 0 0 0;\n",
              "      width: 32px;\n",
              "    }\n",
              "\n",
              "    .colab-df-convert:hover {\n",
              "      background-color: #E2EBFA;\n",
              "      box-shadow: 0px 1px 2px rgba(60, 64, 67, 0.3), 0px 1px 3px 1px rgba(60, 64, 67, 0.15);\n",
              "      fill: #174EA6;\n",
              "    }\n",
              "\n",
              "    [theme=dark] .colab-df-convert {\n",
              "      background-color: #3B4455;\n",
              "      fill: #D2E3FC;\n",
              "    }\n",
              "\n",
              "    [theme=dark] .colab-df-convert:hover {\n",
              "      background-color: #434B5C;\n",
              "      box-shadow: 0px 1px 3px 1px rgba(0, 0, 0, 0.15);\n",
              "      filter: drop-shadow(0px 1px 2px rgba(0, 0, 0, 0.3));\n",
              "      fill: #FFFFFF;\n",
              "    }\n",
              "  </style>\n",
              "\n",
              "      <script>\n",
              "        const buttonEl =\n",
              "          document.querySelector('#df-f933e963-3003-47d9-a977-a891d58130a7 button.colab-df-convert');\n",
              "        buttonEl.style.display =\n",
              "          google.colab.kernel.accessAllowed ? 'block' : 'none';\n",
              "\n",
              "        async function convertToInteractive(key) {\n",
              "          const element = document.querySelector('#df-f933e963-3003-47d9-a977-a891d58130a7');\n",
              "          const dataTable =\n",
              "            await google.colab.kernel.invokeFunction('convertToInteractive',\n",
              "                                                     [key], {});\n",
              "          if (!dataTable) return;\n",
              "\n",
              "          const docLinkHtml = 'Like what you see? Visit the ' +\n",
              "            '<a target=\"_blank\" href=https://colab.research.google.com/notebooks/data_table.ipynb>data table notebook</a>'\n",
              "            + ' to learn more about interactive tables.';\n",
              "          element.innerHTML = '';\n",
              "          dataTable['output_type'] = 'display_data';\n",
              "          await google.colab.output.renderOutput(dataTable, element);\n",
              "          const docLink = document.createElement('div');\n",
              "          docLink.innerHTML = docLinkHtml;\n",
              "          element.appendChild(docLink);\n",
              "        }\n",
              "      </script>\n",
              "    </div>\n",
              "  </div>\n",
              "  "
            ]
          },
          "metadata": {},
          "execution_count": 7
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "Let's add a badwords count to the dataframe."
      ],
      "metadata": {
        "id": "EdKoiwELNYG9"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "with open('/content/drive/MyDrive/Python_data/badwords.txt') as f:\n",
        "    lines = f.readline()\n",
        "\n",
        "badwords = lines.split(',')\n",
        "\n",
        "def badwords_count(text):\n",
        "    try:\n",
        "        text = text.split()\n",
        "        return len([word for word in text if word in badwords])\n",
        "    except:\n",
        "        return 0\n",
        "\n",
        "df['badwords'] = df['processed'].apply(lambda x: badwords_count(x))"
      ],
      "metadata": {
        "id": "pTMZSMoENgSB"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "source": [
        "We drop a 'date' column just for the convenience of printing all informative columns in df.head()"
      ],
      "metadata": {
        "id": "qKU4I7AWYJG-"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "df.drop(['date'], inplace = True, axis = 1)"
      ],
      "metadata": {
        "id": "XDLn7x_HNuMQ"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "df.head()"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 204
        },
        "id": "6xhF0BXZOMdF",
        "outputId": "8064fe42-8a46-49ec-94f2-9b0fc6258f72"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "                                                text           user   num  at  \\\n",
              "0  @SenatorWicker @CNN @jimsciutto Please send 20...   ArtemiyMoroz   105   3   \n",
              "1  @Der_Parrot If anything, I see vast potential ...   AdrianEMoser     0   1   \n",
              "2  the whos feeding the sheep, this probably goes...       werchenz   107   0   \n",
              "3  look you too can support Nazis in Ukraine\\njus...     vandman777  2968   0   \n",
              "4  Create a false pandemic instilling fear &amp; ...  thelastpers0n    37   0   \n",
              "\n",
              "   hash  len  links  expert  \\\n",
              "0     1   80      0       0   \n",
              "1     0  287      0       0   \n",
              "2     0  270      0       0   \n",
              "3     0  110      1       0   \n",
              "4     0  307      1       0   \n",
              "\n",
              "                                           processed  badwords  \n",
              "0        please send more himars to ukraine urgently         0  \n",
              "1  if anything i see vast potential in creating j...         0  \n",
              "2  the whos feeding the sheep this probably goes ...         0  \n",
              "3  look you too can support nazis in ukraine just...         0  \n",
              "4  create a false pandemic instilling fear confor...         0  "
            ],
            "text/html": [
              "\n",
              "  <div id=\"df-4b50b219-2fb0-4ddb-aeae-83c33672f2f8\">\n",
              "    <div class=\"colab-df-container\">\n",
              "      <div>\n",
              "<style scoped>\n",
              "    .dataframe tbody tr th:only-of-type {\n",
              "        vertical-align: middle;\n",
              "    }\n",
              "\n",
              "    .dataframe tbody tr th {\n",
              "        vertical-align: top;\n",
              "    }\n",
              "\n",
              "    .dataframe thead th {\n",
              "        text-align: right;\n",
              "    }\n",
              "</style>\n",
              "<table border=\"1\" class=\"dataframe\">\n",
              "  <thead>\n",
              "    <tr style=\"text-align: right;\">\n",
              "      <th></th>\n",
              "      <th>text</th>\n",
              "      <th>user</th>\n",
              "      <th>num</th>\n",
              "      <th>at</th>\n",
              "      <th>hash</th>\n",
              "      <th>len</th>\n",
              "      <th>links</th>\n",
              "      <th>expert</th>\n",
              "      <th>processed</th>\n",
              "      <th>badwords</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>0</th>\n",
              "      <td>@SenatorWicker @CNN @jimsciutto Please send 20...</td>\n",
              "      <td>ArtemiyMoroz</td>\n",
              "      <td>105</td>\n",
              "      <td>3</td>\n",
              "      <td>1</td>\n",
              "      <td>80</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>please send more himars to ukraine urgently</td>\n",
              "      <td>0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1</th>\n",
              "      <td>@Der_Parrot If anything, I see vast potential ...</td>\n",
              "      <td>AdrianEMoser</td>\n",
              "      <td>0</td>\n",
              "      <td>1</td>\n",
              "      <td>0</td>\n",
              "      <td>287</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>if anything i see vast potential in creating j...</td>\n",
              "      <td>0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>2</th>\n",
              "      <td>the whos feeding the sheep, this probably goes...</td>\n",
              "      <td>werchenz</td>\n",
              "      <td>107</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>270</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>the whos feeding the sheep this probably goes ...</td>\n",
              "      <td>0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>3</th>\n",
              "      <td>look you too can support Nazis in Ukraine\\njus...</td>\n",
              "      <td>vandman777</td>\n",
              "      <td>2968</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>110</td>\n",
              "      <td>1</td>\n",
              "      <td>0</td>\n",
              "      <td>look you too can support nazis in ukraine just...</td>\n",
              "      <td>0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>4</th>\n",
              "      <td>Create a false pandemic instilling fear &amp;amp; ...</td>\n",
              "      <td>thelastpers0n</td>\n",
              "      <td>37</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>307</td>\n",
              "      <td>1</td>\n",
              "      <td>0</td>\n",
              "      <td>create a false pandemic instilling fear confor...</td>\n",
              "      <td>0</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "</div>\n",
              "      <button class=\"colab-df-convert\" onclick=\"convertToInteractive('df-4b50b219-2fb0-4ddb-aeae-83c33672f2f8')\"\n",
              "              title=\"Convert this dataframe to an interactive table.\"\n",
              "              style=\"display:none;\">\n",
              "        \n",
              "  <svg xmlns=\"http://www.w3.org/2000/svg\" height=\"24px\"viewBox=\"0 0 24 24\"\n",
              "       width=\"24px\">\n",
              "    <path d=\"M0 0h24v24H0V0z\" fill=\"none\"/>\n",
              "    <path d=\"M18.56 5.44l.94 2.06.94-2.06 2.06-.94-2.06-.94-.94-2.06-.94 2.06-2.06.94zm-11 1L8.5 8.5l.94-2.06 2.06-.94-2.06-.94L8.5 2.5l-.94 2.06-2.06.94zm10 10l.94 2.06.94-2.06 2.06-.94-2.06-.94-.94-2.06-.94 2.06-2.06.94z\"/><path d=\"M17.41 7.96l-1.37-1.37c-.4-.4-.92-.59-1.43-.59-.52 0-1.04.2-1.43.59L10.3 9.45l-7.72 7.72c-.78.78-.78 2.05 0 2.83L4 21.41c.39.39.9.59 1.41.59.51 0 1.02-.2 1.41-.59l7.78-7.78 2.81-2.81c.8-.78.8-2.07 0-2.86zM5.41 20L4 18.59l7.72-7.72 1.47 1.35L5.41 20z\"/>\n",
              "  </svg>\n",
              "      </button>\n",
              "      \n",
              "  <style>\n",
              "    .colab-df-container {\n",
              "      display:flex;\n",
              "      flex-wrap:wrap;\n",
              "      gap: 12px;\n",
              "    }\n",
              "\n",
              "    .colab-df-convert {\n",
              "      background-color: #E8F0FE;\n",
              "      border: none;\n",
              "      border-radius: 50%;\n",
              "      cursor: pointer;\n",
              "      display: none;\n",
              "      fill: #1967D2;\n",
              "      height: 32px;\n",
              "      padding: 0 0 0 0;\n",
              "      width: 32px;\n",
              "    }\n",
              "\n",
              "    .colab-df-convert:hover {\n",
              "      background-color: #E2EBFA;\n",
              "      box-shadow: 0px 1px 2px rgba(60, 64, 67, 0.3), 0px 1px 3px 1px rgba(60, 64, 67, 0.15);\n",
              "      fill: #174EA6;\n",
              "    }\n",
              "\n",
              "    [theme=dark] .colab-df-convert {\n",
              "      background-color: #3B4455;\n",
              "      fill: #D2E3FC;\n",
              "    }\n",
              "\n",
              "    [theme=dark] .colab-df-convert:hover {\n",
              "      background-color: #434B5C;\n",
              "      box-shadow: 0px 1px 3px 1px rgba(0, 0, 0, 0.15);\n",
              "      filter: drop-shadow(0px 1px 2px rgba(0, 0, 0, 0.3));\n",
              "      fill: #FFFFFF;\n",
              "    }\n",
              "  </style>\n",
              "\n",
              "      <script>\n",
              "        const buttonEl =\n",
              "          document.querySelector('#df-4b50b219-2fb0-4ddb-aeae-83c33672f2f8 button.colab-df-convert');\n",
              "        buttonEl.style.display =\n",
              "          google.colab.kernel.accessAllowed ? 'block' : 'none';\n",
              "\n",
              "        async function convertToInteractive(key) {\n",
              "          const element = document.querySelector('#df-4b50b219-2fb0-4ddb-aeae-83c33672f2f8');\n",
              "          const dataTable =\n",
              "            await google.colab.kernel.invokeFunction('convertToInteractive',\n",
              "                                                     [key], {});\n",
              "          if (!dataTable) return;\n",
              "\n",
              "          const docLinkHtml = 'Like what you see? Visit the ' +\n",
              "            '<a target=\"_blank\" href=https://colab.research.google.com/notebooks/data_table.ipynb>data table notebook</a>'\n",
              "            + ' to learn more about interactive tables.';\n",
              "          element.innerHTML = '';\n",
              "          dataTable['output_type'] = 'display_data';\n",
              "          await google.colab.output.renderOutput(dataTable, element);\n",
              "          const docLink = document.createElement('div');\n",
              "          docLink.innerHTML = docLinkHtml;\n",
              "          element.appendChild(docLink);\n",
              "        }\n",
              "      </script>\n",
              "    </div>\n",
              "  </div>\n",
              "  "
            ]
          },
          "metadata": {},
          "execution_count": 10
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "At this point, I made an assumption that some categorization of the 'at count', 'hash count', 'links count' and 'badwords count' values would make a model less noisy.\n",
        "\n",
        "All values exceeding 3 in these columns were cut to 3. A number of tweets with more than 3 hashes or badwords is extremely low and we can treat these values as outliers. Not to make this notebook too long, I will not dive deeper into this to prove it makes sense."
      ],
      "metadata": {
        "id": "y__h8zzULfTt"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "df['at'] = df['at'].apply(lambda x: 3 if x >=3 else x)\n",
        "df['hash'] = df['hash'].apply(lambda x: 3 if x >=3 else x)\n",
        "df['links'] = df['links'].apply(lambda x: 3 if x >=3 else x)\n",
        "df['badwords'] = df['badwords'].apply(lambda x: 3 if x >=3 else x)"
      ],
      "metadata": {
        "id": "s-RHa8iFO4QG"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "source": [
        "And some final check for the integrity of the dataset to avoid any indexing issues in the following steps."
      ],
      "metadata": {
        "id": "rdGAvu_dPAVB"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "print(df.isna().sum())\n",
        "df.drop_duplicates(inplace = True)\n",
        "df.dropna(inplace = True)\n",
        "df.reset_index(inplace = True, drop = True)\n",
        "print(f'Number of unique, non-null records: {df.shape[0]}.')"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "XNqma5fSPKgD",
        "outputId": "ed6ac7e7-ab1f-47d7-8b3a-472402412e0d"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "text         0\n",
            "user         0\n",
            "num          0\n",
            "at           0\n",
            "hash         0\n",
            "len          0\n",
            "links        0\n",
            "expert       0\n",
            "processed    0\n",
            "badwords     0\n",
            "dtype: int64\n",
            "Number of unique, non-null records: 99732.\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "The difference in a number of records (99.992 after loading a dataset and 99.732 after its cleaning) comes from the fact, that we dropped duplicates after dropping a \"date\" column. It looks like 260 tweets were resent by the same user with the same text body at different date. They seemed \"unique\" only as long as a \"date\" column was in place."
      ],
      "metadata": {
        "id": "fxrR5WeeQmYw"
      }
    },
    {
      "cell_type": "markdown",
      "source": [
        "Now, as the processing is finished, let's save a processed file for the future use."
      ],
      "metadata": {
        "id": "OE1DqS4UTFt9"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "df.to_csv('/content/drive/MyDrive/Python_data/processed_tweets.csv', index = False)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 164
        },
        "id": "QahQ60e5TEhz",
        "outputId": "b312eff4-ef36-420f-c899-bbd13727685d"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "error",
          "ename": "NameError",
          "evalue": "ignored",
          "traceback": [
            "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
            "\u001b[0;31mNameError\u001b[0m                                 Traceback (most recent call last)",
            "\u001b[0;32m<ipython-input-1-cc3594170be3>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m\u001b[0m\n\u001b[0;32m----> 1\u001b[0;31m \u001b[0mdf\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mto_csv\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m'/content/drive/MyDrive/Python_data/processed_tweets.csv'\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mindex\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0;32mFalse\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m",
            "\u001b[0;31mNameError\u001b[0m: name 'df' is not defined"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "df.dropna(inplace = True)\n",
        "df.reset_index(inplace = True)"
      ],
      "metadata": {
        "id": "Hr8SfM33Uw1z"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "df.isna().sum()"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "H0qJdzZ0UrEI",
        "outputId": "48507b86-ffcb-48df-ea2b-36f625a9fd93"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "index        0\n",
              "text         0\n",
              "user         0\n",
              "num          0\n",
              "at           0\n",
              "hash         0\n",
              "len          0\n",
              "links        0\n",
              "expert       0\n",
              "processed    0\n",
              "badwords     0\n",
              "dtype: int64"
            ]
          },
          "metadata": {},
          "execution_count": 15
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "import pandas as pd\n",
        "import numpy as np"
      ],
      "metadata": {
        "id": "MwENxvilULZx"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "df = pd.read_csv('/content/drive/MyDrive/Python_data/processed_tweets.csv')\n",
        "df.dropna(inplace = True)\n",
        "df.reset_index(inplace = True)\n",
        "print(df.shape)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "5tA1mvUaUAQt",
        "outputId": "8accfba1-74c8-40a3-aae8-21ca0c6d2ee7"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "(99731, 11)\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "df.head()"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 204
        },
        "id": "sUuBULfEUVMQ",
        "outputId": "ab8aed89-d897-412b-b915-5c2d8f6eb5e9"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "   index                                               text           user  \\\n",
              "0      0  @SenatorWicker @CNN @jimsciutto Please send 20...   ArtemiyMoroz   \n",
              "1      1  @Der_Parrot If anything, I see vast potential ...   AdrianEMoser   \n",
              "2      2  the whos feeding the sheep, this probably goes...       werchenz   \n",
              "3      3  look you too can support Nazis in Ukraine\\njus...     vandman777   \n",
              "4      4  Create a false pandemic instilling fear &amp; ...  thelastpers0n   \n",
              "\n",
              "    num  at  hash  len  links  expert  \\\n",
              "0   105   3     1   80      0       0   \n",
              "1     0   1     0  287      0       0   \n",
              "2   107   0     0  270      0       0   \n",
              "3  2968   0     0  110      1       0   \n",
              "4    37   0     0  307      1       0   \n",
              "\n",
              "                                           processed  badwords  \n",
              "0        please send more himars to ukraine urgently         0  \n",
              "1  if anything i see vast potential in creating j...         0  \n",
              "2  the whos feeding the sheep this probably goes ...         0  \n",
              "3  look you too can support nazis in ukraine just...         0  \n",
              "4  create a false pandemic instilling fear confor...         0  "
            ],
            "text/html": [
              "\n",
              "  <div id=\"df-58f3ec60-f424-4efd-b2e7-f7eb73f9603a\">\n",
              "    <div class=\"colab-df-container\">\n",
              "      <div>\n",
              "<style scoped>\n",
              "    .dataframe tbody tr th:only-of-type {\n",
              "        vertical-align: middle;\n",
              "    }\n",
              "\n",
              "    .dataframe tbody tr th {\n",
              "        vertical-align: top;\n",
              "    }\n",
              "\n",
              "    .dataframe thead th {\n",
              "        text-align: right;\n",
              "    }\n",
              "</style>\n",
              "<table border=\"1\" class=\"dataframe\">\n",
              "  <thead>\n",
              "    <tr style=\"text-align: right;\">\n",
              "      <th></th>\n",
              "      <th>index</th>\n",
              "      <th>text</th>\n",
              "      <th>user</th>\n",
              "      <th>num</th>\n",
              "      <th>at</th>\n",
              "      <th>hash</th>\n",
              "      <th>len</th>\n",
              "      <th>links</th>\n",
              "      <th>expert</th>\n",
              "      <th>processed</th>\n",
              "      <th>badwords</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>0</th>\n",
              "      <td>0</td>\n",
              "      <td>@SenatorWicker @CNN @jimsciutto Please send 20...</td>\n",
              "      <td>ArtemiyMoroz</td>\n",
              "      <td>105</td>\n",
              "      <td>3</td>\n",
              "      <td>1</td>\n",
              "      <td>80</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>please send more himars to ukraine urgently</td>\n",
              "      <td>0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1</th>\n",
              "      <td>1</td>\n",
              "      <td>@Der_Parrot If anything, I see vast potential ...</td>\n",
              "      <td>AdrianEMoser</td>\n",
              "      <td>0</td>\n",
              "      <td>1</td>\n",
              "      <td>0</td>\n",
              "      <td>287</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>if anything i see vast potential in creating j...</td>\n",
              "      <td>0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>2</th>\n",
              "      <td>2</td>\n",
              "      <td>the whos feeding the sheep, this probably goes...</td>\n",
              "      <td>werchenz</td>\n",
              "      <td>107</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>270</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>the whos feeding the sheep this probably goes ...</td>\n",
              "      <td>0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>3</th>\n",
              "      <td>3</td>\n",
              "      <td>look you too can support Nazis in Ukraine\\njus...</td>\n",
              "      <td>vandman777</td>\n",
              "      <td>2968</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>110</td>\n",
              "      <td>1</td>\n",
              "      <td>0</td>\n",
              "      <td>look you too can support nazis in ukraine just...</td>\n",
              "      <td>0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>4</th>\n",
              "      <td>4</td>\n",
              "      <td>Create a false pandemic instilling fear &amp;amp; ...</td>\n",
              "      <td>thelastpers0n</td>\n",
              "      <td>37</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>307</td>\n",
              "      <td>1</td>\n",
              "      <td>0</td>\n",
              "      <td>create a false pandemic instilling fear confor...</td>\n",
              "      <td>0</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "</div>\n",
              "      <button class=\"colab-df-convert\" onclick=\"convertToInteractive('df-58f3ec60-f424-4efd-b2e7-f7eb73f9603a')\"\n",
              "              title=\"Convert this dataframe to an interactive table.\"\n",
              "              style=\"display:none;\">\n",
              "        \n",
              "  <svg xmlns=\"http://www.w3.org/2000/svg\" height=\"24px\"viewBox=\"0 0 24 24\"\n",
              "       width=\"24px\">\n",
              "    <path d=\"M0 0h24v24H0V0z\" fill=\"none\"/>\n",
              "    <path d=\"M18.56 5.44l.94 2.06.94-2.06 2.06-.94-2.06-.94-.94-2.06-.94 2.06-2.06.94zm-11 1L8.5 8.5l.94-2.06 2.06-.94-2.06-.94L8.5 2.5l-.94 2.06-2.06.94zm10 10l.94 2.06.94-2.06 2.06-.94-2.06-.94-.94-2.06-.94 2.06-2.06.94z\"/><path d=\"M17.41 7.96l-1.37-1.37c-.4-.4-.92-.59-1.43-.59-.52 0-1.04.2-1.43.59L10.3 9.45l-7.72 7.72c-.78.78-.78 2.05 0 2.83L4 21.41c.39.39.9.59 1.41.59.51 0 1.02-.2 1.41-.59l7.78-7.78 2.81-2.81c.8-.78.8-2.07 0-2.86zM5.41 20L4 18.59l7.72-7.72 1.47 1.35L5.41 20z\"/>\n",
              "  </svg>\n",
              "      </button>\n",
              "      \n",
              "  <style>\n",
              "    .colab-df-container {\n",
              "      display:flex;\n",
              "      flex-wrap:wrap;\n",
              "      gap: 12px;\n",
              "    }\n",
              "\n",
              "    .colab-df-convert {\n",
              "      background-color: #E8F0FE;\n",
              "      border: none;\n",
              "      border-radius: 50%;\n",
              "      cursor: pointer;\n",
              "      display: none;\n",
              "      fill: #1967D2;\n",
              "      height: 32px;\n",
              "      padding: 0 0 0 0;\n",
              "      width: 32px;\n",
              "    }\n",
              "\n",
              "    .colab-df-convert:hover {\n",
              "      background-color: #E2EBFA;\n",
              "      box-shadow: 0px 1px 2px rgba(60, 64, 67, 0.3), 0px 1px 3px 1px rgba(60, 64, 67, 0.15);\n",
              "      fill: #174EA6;\n",
              "    }\n",
              "\n",
              "    [theme=dark] .colab-df-convert {\n",
              "      background-color: #3B4455;\n",
              "      fill: #D2E3FC;\n",
              "    }\n",
              "\n",
              "    [theme=dark] .colab-df-convert:hover {\n",
              "      background-color: #434B5C;\n",
              "      box-shadow: 0px 1px 3px 1px rgba(0, 0, 0, 0.15);\n",
              "      filter: drop-shadow(0px 1px 2px rgba(0, 0, 0, 0.3));\n",
              "      fill: #FFFFFF;\n",
              "    }\n",
              "  </style>\n",
              "\n",
              "      <script>\n",
              "        const buttonEl =\n",
              "          document.querySelector('#df-58f3ec60-f424-4efd-b2e7-f7eb73f9603a button.colab-df-convert');\n",
              "        buttonEl.style.display =\n",
              "          google.colab.kernel.accessAllowed ? 'block' : 'none';\n",
              "\n",
              "        async function convertToInteractive(key) {\n",
              "          const element = document.querySelector('#df-58f3ec60-f424-4efd-b2e7-f7eb73f9603a');\n",
              "          const dataTable =\n",
              "            await google.colab.kernel.invokeFunction('convertToInteractive',\n",
              "                                                     [key], {});\n",
              "          if (!dataTable) return;\n",
              "\n",
              "          const docLinkHtml = 'Like what you see? Visit the ' +\n",
              "            '<a target=\"_blank\" href=https://colab.research.google.com/notebooks/data_table.ipynb>data table notebook</a>'\n",
              "            + ' to learn more about interactive tables.';\n",
              "          element.innerHTML = '';\n",
              "          dataTable['output_type'] = 'display_data';\n",
              "          await google.colab.output.renderOutput(dataTable, element);\n",
              "          const docLink = document.createElement('div');\n",
              "          docLink.innerHTML = docLinkHtml;\n",
              "          element.appendChild(docLink);\n",
              "        }\n",
              "      </script>\n",
              "    </div>\n",
              "  </div>\n",
              "  "
            ]
          },
          "metadata": {},
          "execution_count": 4
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "**CountVectorization**\n",
        "Count vectorization usually produces weaker resuluts than TF-IDF vectorization, but we'll use Count vectorizer to make some analysis on which words occure more often in \"expert\" and \"non-expert\" tweets. That should tell us a little bit about a nature of our \"project\"."
      ],
      "metadata": {
        "id": "xda1qs7KSYJZ"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "from sklearn.feature_extraction.text import CountVectorizer"
      ],
      "metadata": {
        "id": "79bGL2Fum6wn"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "source": [
        "Below a smart snipped of a code that I copied from some medium.com article. Does the job to create a DataFrame with each word count separately for \"expert\" and \"non expert\" words."
      ],
      "metadata": {
        "id": "LZvrOdeKoI3M"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "cv = CountVectorizer(stop_words = 'english')\n",
        "cv.fit(df['processed'])\n",
        "\n",
        "cv_pos = cv.transform(df[df['expert'] == 1]['processed'])\n",
        "cv_neg = cv.transform(df[df['expert'] == 0]['processed'])\n",
        "\n",
        "pos_tf = np.sum(cv_pos,axis=0)\n",
        "neg_tf = np.sum(cv_neg,axis=0)\n",
        "neg = np.squeeze(np.asarray(neg_tf))\n",
        "pos = np.squeeze(np.asarray(pos_tf))\n",
        "freq = pd.DataFrame([neg,pos],columns=cv.get_feature_names_out()).transpose()\n",
        "freq.columns = 'not expert'.split() # column name 'not' for a non-expert, just to make it shorter"
      ],
      "metadata": {
        "id": "kLA4AtF6nyK_"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "freq.head(10)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 359
        },
        "id": "TZaXW9Jwph8x",
        "outputId": "26a99060-25ca-406b-8177-266258ed0bd5"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "           not  expert\n",
              "aa          23       1\n",
              "aaa          3       0\n",
              "aaaa         2       0\n",
              "aaaaaaagh    1       0\n",
              "aaaand       2       0\n",
              "aaah         1       0\n",
              "aaand        2       0\n",
              "aahhh        2       0\n",
              "aam          1       0\n",
              "aand         1       0"
            ],
            "text/html": [
              "\n",
              "  <div id=\"df-04aaf8e2-277c-4bd7-88c1-a6ef0d54728c\">\n",
              "    <div class=\"colab-df-container\">\n",
              "      <div>\n",
              "<style scoped>\n",
              "    .dataframe tbody tr th:only-of-type {\n",
              "        vertical-align: middle;\n",
              "    }\n",
              "\n",
              "    .dataframe tbody tr th {\n",
              "        vertical-align: top;\n",
              "    }\n",
              "\n",
              "    .dataframe thead th {\n",
              "        text-align: right;\n",
              "    }\n",
              "</style>\n",
              "<table border=\"1\" class=\"dataframe\">\n",
              "  <thead>\n",
              "    <tr style=\"text-align: right;\">\n",
              "      <th></th>\n",
              "      <th>not</th>\n",
              "      <th>expert</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>aa</th>\n",
              "      <td>23</td>\n",
              "      <td>1</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>aaa</th>\n",
              "      <td>3</td>\n",
              "      <td>0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>aaaa</th>\n",
              "      <td>2</td>\n",
              "      <td>0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>aaaaaaagh</th>\n",
              "      <td>1</td>\n",
              "      <td>0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>aaaand</th>\n",
              "      <td>2</td>\n",
              "      <td>0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>aaah</th>\n",
              "      <td>1</td>\n",
              "      <td>0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>aaand</th>\n",
              "      <td>2</td>\n",
              "      <td>0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>aahhh</th>\n",
              "      <td>2</td>\n",
              "      <td>0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>aam</th>\n",
              "      <td>1</td>\n",
              "      <td>0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>aand</th>\n",
              "      <td>1</td>\n",
              "      <td>0</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "</div>\n",
              "      <button class=\"colab-df-convert\" onclick=\"convertToInteractive('df-04aaf8e2-277c-4bd7-88c1-a6ef0d54728c')\"\n",
              "              title=\"Convert this dataframe to an interactive table.\"\n",
              "              style=\"display:none;\">\n",
              "        \n",
              "  <svg xmlns=\"http://www.w3.org/2000/svg\" height=\"24px\"viewBox=\"0 0 24 24\"\n",
              "       width=\"24px\">\n",
              "    <path d=\"M0 0h24v24H0V0z\" fill=\"none\"/>\n",
              "    <path d=\"M18.56 5.44l.94 2.06.94-2.06 2.06-.94-2.06-.94-.94-2.06-.94 2.06-2.06.94zm-11 1L8.5 8.5l.94-2.06 2.06-.94-2.06-.94L8.5 2.5l-.94 2.06-2.06.94zm10 10l.94 2.06.94-2.06 2.06-.94-2.06-.94-.94-2.06-.94 2.06-2.06.94z\"/><path d=\"M17.41 7.96l-1.37-1.37c-.4-.4-.92-.59-1.43-.59-.52 0-1.04.2-1.43.59L10.3 9.45l-7.72 7.72c-.78.78-.78 2.05 0 2.83L4 21.41c.39.39.9.59 1.41.59.51 0 1.02-.2 1.41-.59l7.78-7.78 2.81-2.81c.8-.78.8-2.07 0-2.86zM5.41 20L4 18.59l7.72-7.72 1.47 1.35L5.41 20z\"/>\n",
              "  </svg>\n",
              "      </button>\n",
              "      \n",
              "  <style>\n",
              "    .colab-df-container {\n",
              "      display:flex;\n",
              "      flex-wrap:wrap;\n",
              "      gap: 12px;\n",
              "    }\n",
              "\n",
              "    .colab-df-convert {\n",
              "      background-color: #E8F0FE;\n",
              "      border: none;\n",
              "      border-radius: 50%;\n",
              "      cursor: pointer;\n",
              "      display: none;\n",
              "      fill: #1967D2;\n",
              "      height: 32px;\n",
              "      padding: 0 0 0 0;\n",
              "      width: 32px;\n",
              "    }\n",
              "\n",
              "    .colab-df-convert:hover {\n",
              "      background-color: #E2EBFA;\n",
              "      box-shadow: 0px 1px 2px rgba(60, 64, 67, 0.3), 0px 1px 3px 1px rgba(60, 64, 67, 0.15);\n",
              "      fill: #174EA6;\n",
              "    }\n",
              "\n",
              "    [theme=dark] .colab-df-convert {\n",
              "      background-color: #3B4455;\n",
              "      fill: #D2E3FC;\n",
              "    }\n",
              "\n",
              "    [theme=dark] .colab-df-convert:hover {\n",
              "      background-color: #434B5C;\n",
              "      box-shadow: 0px 1px 3px 1px rgba(0, 0, 0, 0.15);\n",
              "      filter: drop-shadow(0px 1px 2px rgba(0, 0, 0, 0.3));\n",
              "      fill: #FFFFFF;\n",
              "    }\n",
              "  </style>\n",
              "\n",
              "      <script>\n",
              "        const buttonEl =\n",
              "          document.querySelector('#df-04aaf8e2-277c-4bd7-88c1-a6ef0d54728c button.colab-df-convert');\n",
              "        buttonEl.style.display =\n",
              "          google.colab.kernel.accessAllowed ? 'block' : 'none';\n",
              "\n",
              "        async function convertToInteractive(key) {\n",
              "          const element = document.querySelector('#df-04aaf8e2-277c-4bd7-88c1-a6ef0d54728c');\n",
              "          const dataTable =\n",
              "            await google.colab.kernel.invokeFunction('convertToInteractive',\n",
              "                                                     [key], {});\n",
              "          if (!dataTable) return;\n",
              "\n",
              "          const docLinkHtml = 'Like what you see? Visit the ' +\n",
              "            '<a target=\"_blank\" href=https://colab.research.google.com/notebooks/data_table.ipynb>data table notebook</a>'\n",
              "            + ' to learn more about interactive tables.';\n",
              "          element.innerHTML = '';\n",
              "          dataTable['output_type'] = 'display_data';\n",
              "          await google.colab.output.renderOutput(dataTable, element);\n",
              "          const docLink = document.createElement('div');\n",
              "          docLink.innerHTML = docLinkHtml;\n",
              "          element.appendChild(docLink);\n",
              "        }\n",
              "      </script>\n",
              "    </div>\n",
              "  </div>\n",
              "  "
            ]
          },
          "metadata": {},
          "execution_count": 10
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "Before we begin to compare both columns, let's do a couple of simple data manipulations:\n",
        "\n",
        "\n",
        "1.   Let's bring the values of a \"not expert\" column to the same base as an \"expert\" column by dividng these values by a denominator len(expert) / len(not expert). It will make comparisions more intuitive. I also rounded the values.\n",
        "2.   Let's add a 'total' count\n",
        "3. I also removed rows with a 'total' equal to zero (low 'not expert' values after dividing by a denominator and rounding were flattened to zero, which doesn't really matter - these were words of marginal analitic worth, like 'aa' and similar \n",
        "\n"
      ],
      "metadata": {
        "id": "nW_cpPNRpn2X"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "denominator = len(df[df['expert'] == 1]) / len(df[df['expert'] == 0])\n",
        "\n",
        "freq['not'] = (freq['not'] * denominator).round().astype('int32')\n",
        "freq['total'] = freq.sum(axis = 1)\n",
        "freq.drop(freq[freq['total'] == 0].index, inplace = True) # pozbywamy się tych, gdzie total = 0"
      ],
      "metadata": {
        "id": "rdFmZwP7ruhQ"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "freq.sort_values('total', ascending = False).head(20)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 669
        },
        "id": "a2g75kdPr1Lx",
        "outputId": "6f9dfba3-51fa-43ba-c3d6-ba0b0d82963d"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "            not  expert  total\n",
              "ukraine    6882    6865  13747\n",
              "russia     1863    2361   4224\n",
              "war        1156    1371   2527\n",
              "russian     866    1309   2175\n",
              "putin       765    1029   1794\n",
              "people      510     323    833\n",
              "ukrainian   283     464    747\n",
              "just        463     278    741\n",
              "world       276     438    714\n",
              "like        430     266    696\n",
              "biden       323     333    656\n",
              "military    270     374    644\n",
              "news        179     412    591\n",
              "support     295     274    569\n",
              "china       191     369    560\n",
              "nuclear     192     356    548\n",
              "money       376     156    532\n",
              "president   165     362    527\n",
              "says        128     379    507\n",
              "invasion    190     315    505"
            ],
            "text/html": [
              "\n",
              "  <div id=\"df-68a3f2eb-05c7-4a5f-8733-a310f140aec3\">\n",
              "    <div class=\"colab-df-container\">\n",
              "      <div>\n",
              "<style scoped>\n",
              "    .dataframe tbody tr th:only-of-type {\n",
              "        vertical-align: middle;\n",
              "    }\n",
              "\n",
              "    .dataframe tbody tr th {\n",
              "        vertical-align: top;\n",
              "    }\n",
              "\n",
              "    .dataframe thead th {\n",
              "        text-align: right;\n",
              "    }\n",
              "</style>\n",
              "<table border=\"1\" class=\"dataframe\">\n",
              "  <thead>\n",
              "    <tr style=\"text-align: right;\">\n",
              "      <th></th>\n",
              "      <th>not</th>\n",
              "      <th>expert</th>\n",
              "      <th>total</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>ukraine</th>\n",
              "      <td>6882</td>\n",
              "      <td>6865</td>\n",
              "      <td>13747</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>russia</th>\n",
              "      <td>1863</td>\n",
              "      <td>2361</td>\n",
              "      <td>4224</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>war</th>\n",
              "      <td>1156</td>\n",
              "      <td>1371</td>\n",
              "      <td>2527</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>russian</th>\n",
              "      <td>866</td>\n",
              "      <td>1309</td>\n",
              "      <td>2175</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>putin</th>\n",
              "      <td>765</td>\n",
              "      <td>1029</td>\n",
              "      <td>1794</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>people</th>\n",
              "      <td>510</td>\n",
              "      <td>323</td>\n",
              "      <td>833</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>ukrainian</th>\n",
              "      <td>283</td>\n",
              "      <td>464</td>\n",
              "      <td>747</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>just</th>\n",
              "      <td>463</td>\n",
              "      <td>278</td>\n",
              "      <td>741</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>world</th>\n",
              "      <td>276</td>\n",
              "      <td>438</td>\n",
              "      <td>714</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>like</th>\n",
              "      <td>430</td>\n",
              "      <td>266</td>\n",
              "      <td>696</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>biden</th>\n",
              "      <td>323</td>\n",
              "      <td>333</td>\n",
              "      <td>656</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>military</th>\n",
              "      <td>270</td>\n",
              "      <td>374</td>\n",
              "      <td>644</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>news</th>\n",
              "      <td>179</td>\n",
              "      <td>412</td>\n",
              "      <td>591</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>support</th>\n",
              "      <td>295</td>\n",
              "      <td>274</td>\n",
              "      <td>569</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>china</th>\n",
              "      <td>191</td>\n",
              "      <td>369</td>\n",
              "      <td>560</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>nuclear</th>\n",
              "      <td>192</td>\n",
              "      <td>356</td>\n",
              "      <td>548</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>money</th>\n",
              "      <td>376</td>\n",
              "      <td>156</td>\n",
              "      <td>532</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>president</th>\n",
              "      <td>165</td>\n",
              "      <td>362</td>\n",
              "      <td>527</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>says</th>\n",
              "      <td>128</td>\n",
              "      <td>379</td>\n",
              "      <td>507</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>invasion</th>\n",
              "      <td>190</td>\n",
              "      <td>315</td>\n",
              "      <td>505</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "</div>\n",
              "      <button class=\"colab-df-convert\" onclick=\"convertToInteractive('df-68a3f2eb-05c7-4a5f-8733-a310f140aec3')\"\n",
              "              title=\"Convert this dataframe to an interactive table.\"\n",
              "              style=\"display:none;\">\n",
              "        \n",
              "  <svg xmlns=\"http://www.w3.org/2000/svg\" height=\"24px\"viewBox=\"0 0 24 24\"\n",
              "       width=\"24px\">\n",
              "    <path d=\"M0 0h24v24H0V0z\" fill=\"none\"/>\n",
              "    <path d=\"M18.56 5.44l.94 2.06.94-2.06 2.06-.94-2.06-.94-.94-2.06-.94 2.06-2.06.94zm-11 1L8.5 8.5l.94-2.06 2.06-.94-2.06-.94L8.5 2.5l-.94 2.06-2.06.94zm10 10l.94 2.06.94-2.06 2.06-.94-2.06-.94-.94-2.06-.94 2.06-2.06.94z\"/><path d=\"M17.41 7.96l-1.37-1.37c-.4-.4-.92-.59-1.43-.59-.52 0-1.04.2-1.43.59L10.3 9.45l-7.72 7.72c-.78.78-.78 2.05 0 2.83L4 21.41c.39.39.9.59 1.41.59.51 0 1.02-.2 1.41-.59l7.78-7.78 2.81-2.81c.8-.78.8-2.07 0-2.86zM5.41 20L4 18.59l7.72-7.72 1.47 1.35L5.41 20z\"/>\n",
              "  </svg>\n",
              "      </button>\n",
              "      \n",
              "  <style>\n",
              "    .colab-df-container {\n",
              "      display:flex;\n",
              "      flex-wrap:wrap;\n",
              "      gap: 12px;\n",
              "    }\n",
              "\n",
              "    .colab-df-convert {\n",
              "      background-color: #E8F0FE;\n",
              "      border: none;\n",
              "      border-radius: 50%;\n",
              "      cursor: pointer;\n",
              "      display: none;\n",
              "      fill: #1967D2;\n",
              "      height: 32px;\n",
              "      padding: 0 0 0 0;\n",
              "      width: 32px;\n",
              "    }\n",
              "\n",
              "    .colab-df-convert:hover {\n",
              "      background-color: #E2EBFA;\n",
              "      box-shadow: 0px 1px 2px rgba(60, 64, 67, 0.3), 0px 1px 3px 1px rgba(60, 64, 67, 0.15);\n",
              "      fill: #174EA6;\n",
              "    }\n",
              "\n",
              "    [theme=dark] .colab-df-convert {\n",
              "      background-color: #3B4455;\n",
              "      fill: #D2E3FC;\n",
              "    }\n",
              "\n",
              "    [theme=dark] .colab-df-convert:hover {\n",
              "      background-color: #434B5C;\n",
              "      box-shadow: 0px 1px 3px 1px rgba(0, 0, 0, 0.15);\n",
              "      filter: drop-shadow(0px 1px 2px rgba(0, 0, 0, 0.3));\n",
              "      fill: #FFFFFF;\n",
              "    }\n",
              "  </style>\n",
              "\n",
              "      <script>\n",
              "        const buttonEl =\n",
              "          document.querySelector('#df-68a3f2eb-05c7-4a5f-8733-a310f140aec3 button.colab-df-convert');\n",
              "        buttonEl.style.display =\n",
              "          google.colab.kernel.accessAllowed ? 'block' : 'none';\n",
              "\n",
              "        async function convertToInteractive(key) {\n",
              "          const element = document.querySelector('#df-68a3f2eb-05c7-4a5f-8733-a310f140aec3');\n",
              "          const dataTable =\n",
              "            await google.colab.kernel.invokeFunction('convertToInteractive',\n",
              "                                                     [key], {});\n",
              "          if (!dataTable) return;\n",
              "\n",
              "          const docLinkHtml = 'Like what you see? Visit the ' +\n",
              "            '<a target=\"_blank\" href=https://colab.research.google.com/notebooks/data_table.ipynb>data table notebook</a>'\n",
              "            + ' to learn more about interactive tables.';\n",
              "          element.innerHTML = '';\n",
              "          dataTable['output_type'] = 'display_data';\n",
              "          await google.colab.output.renderOutput(dataTable, element);\n",
              "          const docLink = document.createElement('div');\n",
              "          docLink.innerHTML = docLinkHtml;\n",
              "          element.appendChild(docLink);\n",
              "        }\n",
              "      </script>\n",
              "    </div>\n",
              "  </div>\n",
              "  "
            ]
          },
          "metadata": {},
          "execution_count": 12
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "As values are denominated to the same base, we can easily compare them. At first sight we see (as an example) that experts refer to Russia much more often, than ordinary users. We also see much more reference to China or nuclear threat."
      ],
      "metadata": {
        "id": "yTFtk60fsXQ_"
      }
    },
    {
      "cell_type": "markdown",
      "source": [
        "And the most interesting part. Let's find words that appear in 'expert' and 'not expert' columns much more often than in the opposite one.\n",
        "To do so, we'll add two columns:\n",
        "'expert_rate' - a share of all word occurances in 'expert' tweets. The higher the value, to more this word is biased towards 'expert' tweets. Value 1.0 means this words occurs only in 'expert' tweets.\n",
        "Let's take a look at top 20 'expert' words. We dropped all words with fewer occurances than 10 in a dataset as irrelevant."
      ],
      "metadata": {
        "id": "tD_z8YURxbBh"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "freq['expert_rate'] = freq['expert'] / freq['total']\n",
        "freq['not_expert_rate'] = freq['not'] / freq['total']"
      ],
      "metadata": {
        "id": "Bz9ALKxavu50"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "freq[freq['total'] > 10].sort_values('expert_rate', ascending = False).head(20)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 669
        },
        "id": "oPyOTXaoyfvX",
        "outputId": "9334fe63-2764-418c-d386-7125527b2201"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "                  not  expert  total  expert_rate  not_expert_rate\n",
              "nr                  0      11     11     1.000000         0.000000\n",
              "wtpblue             0      13     13     1.000000         0.000000\n",
              "vermaut             0      54     54     1.000000         0.000000\n",
              "demvoice            0      17     17     1.000000         0.000000\n",
              "mostshared          0      12     12     1.000000         0.000000\n",
              "mostliked           0      11     11     1.000000         0.000000\n",
              "riskmap             0      17     17     1.000000         0.000000\n",
              "submitted           0      15     15     1.000000         0.000000\n",
              "retrospective       0      18     18     1.000000         0.000000\n",
              "twitters            0      13     13     1.000000         0.000000\n",
              "strongertogether    1     196    197     0.994924         0.005076\n",
              "bb                  2     196    198     0.989899         0.010101\n",
              "shorts              1      65     66     0.984848         0.015152\n",
              "andy                1      61     62     0.983871         0.016129\n",
              "ucrania             1      60     61     0.983607         0.016393\n",
              "cnbc                2      66     68     0.970588         0.029412\n",
              "shares              2      65     67     0.970149         0.029851\n",
              "rusia               2      60     62     0.967742         0.032258\n",
              "nbc                 3      90     93     0.967742         0.032258\n",
              "movie               7     205    212     0.966981         0.033019"
            ],
            "text/html": [
              "\n",
              "  <div id=\"df-cf7ee0da-41aa-44e0-b3c0-5d84a8bfb073\">\n",
              "    <div class=\"colab-df-container\">\n",
              "      <div>\n",
              "<style scoped>\n",
              "    .dataframe tbody tr th:only-of-type {\n",
              "        vertical-align: middle;\n",
              "    }\n",
              "\n",
              "    .dataframe tbody tr th {\n",
              "        vertical-align: top;\n",
              "    }\n",
              "\n",
              "    .dataframe thead th {\n",
              "        text-align: right;\n",
              "    }\n",
              "</style>\n",
              "<table border=\"1\" class=\"dataframe\">\n",
              "  <thead>\n",
              "    <tr style=\"text-align: right;\">\n",
              "      <th></th>\n",
              "      <th>not</th>\n",
              "      <th>expert</th>\n",
              "      <th>total</th>\n",
              "      <th>expert_rate</th>\n",
              "      <th>not_expert_rate</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>nr</th>\n",
              "      <td>0</td>\n",
              "      <td>11</td>\n",
              "      <td>11</td>\n",
              "      <td>1.000000</td>\n",
              "      <td>0.000000</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>wtpblue</th>\n",
              "      <td>0</td>\n",
              "      <td>13</td>\n",
              "      <td>13</td>\n",
              "      <td>1.000000</td>\n",
              "      <td>0.000000</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>vermaut</th>\n",
              "      <td>0</td>\n",
              "      <td>54</td>\n",
              "      <td>54</td>\n",
              "      <td>1.000000</td>\n",
              "      <td>0.000000</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>demvoice</th>\n",
              "      <td>0</td>\n",
              "      <td>17</td>\n",
              "      <td>17</td>\n",
              "      <td>1.000000</td>\n",
              "      <td>0.000000</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>mostshared</th>\n",
              "      <td>0</td>\n",
              "      <td>12</td>\n",
              "      <td>12</td>\n",
              "      <td>1.000000</td>\n",
              "      <td>0.000000</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>mostliked</th>\n",
              "      <td>0</td>\n",
              "      <td>11</td>\n",
              "      <td>11</td>\n",
              "      <td>1.000000</td>\n",
              "      <td>0.000000</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>riskmap</th>\n",
              "      <td>0</td>\n",
              "      <td>17</td>\n",
              "      <td>17</td>\n",
              "      <td>1.000000</td>\n",
              "      <td>0.000000</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>submitted</th>\n",
              "      <td>0</td>\n",
              "      <td>15</td>\n",
              "      <td>15</td>\n",
              "      <td>1.000000</td>\n",
              "      <td>0.000000</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>retrospective</th>\n",
              "      <td>0</td>\n",
              "      <td>18</td>\n",
              "      <td>18</td>\n",
              "      <td>1.000000</td>\n",
              "      <td>0.000000</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>twitters</th>\n",
              "      <td>0</td>\n",
              "      <td>13</td>\n",
              "      <td>13</td>\n",
              "      <td>1.000000</td>\n",
              "      <td>0.000000</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>strongertogether</th>\n",
              "      <td>1</td>\n",
              "      <td>196</td>\n",
              "      <td>197</td>\n",
              "      <td>0.994924</td>\n",
              "      <td>0.005076</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>bb</th>\n",
              "      <td>2</td>\n",
              "      <td>196</td>\n",
              "      <td>198</td>\n",
              "      <td>0.989899</td>\n",
              "      <td>0.010101</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>shorts</th>\n",
              "      <td>1</td>\n",
              "      <td>65</td>\n",
              "      <td>66</td>\n",
              "      <td>0.984848</td>\n",
              "      <td>0.015152</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>andy</th>\n",
              "      <td>1</td>\n",
              "      <td>61</td>\n",
              "      <td>62</td>\n",
              "      <td>0.983871</td>\n",
              "      <td>0.016129</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>ucrania</th>\n",
              "      <td>1</td>\n",
              "      <td>60</td>\n",
              "      <td>61</td>\n",
              "      <td>0.983607</td>\n",
              "      <td>0.016393</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>cnbc</th>\n",
              "      <td>2</td>\n",
              "      <td>66</td>\n",
              "      <td>68</td>\n",
              "      <td>0.970588</td>\n",
              "      <td>0.029412</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>shares</th>\n",
              "      <td>2</td>\n",
              "      <td>65</td>\n",
              "      <td>67</td>\n",
              "      <td>0.970149</td>\n",
              "      <td>0.029851</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>rusia</th>\n",
              "      <td>2</td>\n",
              "      <td>60</td>\n",
              "      <td>62</td>\n",
              "      <td>0.967742</td>\n",
              "      <td>0.032258</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>nbc</th>\n",
              "      <td>3</td>\n",
              "      <td>90</td>\n",
              "      <td>93</td>\n",
              "      <td>0.967742</td>\n",
              "      <td>0.032258</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>movie</th>\n",
              "      <td>7</td>\n",
              "      <td>205</td>\n",
              "      <td>212</td>\n",
              "      <td>0.966981</td>\n",
              "      <td>0.033019</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "</div>\n",
              "      <button class=\"colab-df-convert\" onclick=\"convertToInteractive('df-cf7ee0da-41aa-44e0-b3c0-5d84a8bfb073')\"\n",
              "              title=\"Convert this dataframe to an interactive table.\"\n",
              "              style=\"display:none;\">\n",
              "        \n",
              "  <svg xmlns=\"http://www.w3.org/2000/svg\" height=\"24px\"viewBox=\"0 0 24 24\"\n",
              "       width=\"24px\">\n",
              "    <path d=\"M0 0h24v24H0V0z\" fill=\"none\"/>\n",
              "    <path d=\"M18.56 5.44l.94 2.06.94-2.06 2.06-.94-2.06-.94-.94-2.06-.94 2.06-2.06.94zm-11 1L8.5 8.5l.94-2.06 2.06-.94-2.06-.94L8.5 2.5l-.94 2.06-2.06.94zm10 10l.94 2.06.94-2.06 2.06-.94-2.06-.94-.94-2.06-.94 2.06-2.06.94z\"/><path d=\"M17.41 7.96l-1.37-1.37c-.4-.4-.92-.59-1.43-.59-.52 0-1.04.2-1.43.59L10.3 9.45l-7.72 7.72c-.78.78-.78 2.05 0 2.83L4 21.41c.39.39.9.59 1.41.59.51 0 1.02-.2 1.41-.59l7.78-7.78 2.81-2.81c.8-.78.8-2.07 0-2.86zM5.41 20L4 18.59l7.72-7.72 1.47 1.35L5.41 20z\"/>\n",
              "  </svg>\n",
              "      </button>\n",
              "      \n",
              "  <style>\n",
              "    .colab-df-container {\n",
              "      display:flex;\n",
              "      flex-wrap:wrap;\n",
              "      gap: 12px;\n",
              "    }\n",
              "\n",
              "    .colab-df-convert {\n",
              "      background-color: #E8F0FE;\n",
              "      border: none;\n",
              "      border-radius: 50%;\n",
              "      cursor: pointer;\n",
              "      display: none;\n",
              "      fill: #1967D2;\n",
              "      height: 32px;\n",
              "      padding: 0 0 0 0;\n",
              "      width: 32px;\n",
              "    }\n",
              "\n",
              "    .colab-df-convert:hover {\n",
              "      background-color: #E2EBFA;\n",
              "      box-shadow: 0px 1px 2px rgba(60, 64, 67, 0.3), 0px 1px 3px 1px rgba(60, 64, 67, 0.15);\n",
              "      fill: #174EA6;\n",
              "    }\n",
              "\n",
              "    [theme=dark] .colab-df-convert {\n",
              "      background-color: #3B4455;\n",
              "      fill: #D2E3FC;\n",
              "    }\n",
              "\n",
              "    [theme=dark] .colab-df-convert:hover {\n",
              "      background-color: #434B5C;\n",
              "      box-shadow: 0px 1px 3px 1px rgba(0, 0, 0, 0.15);\n",
              "      filter: drop-shadow(0px 1px 2px rgba(0, 0, 0, 0.3));\n",
              "      fill: #FFFFFF;\n",
              "    }\n",
              "  </style>\n",
              "\n",
              "      <script>\n",
              "        const buttonEl =\n",
              "          document.querySelector('#df-cf7ee0da-41aa-44e0-b3c0-5d84a8bfb073 button.colab-df-convert');\n",
              "        buttonEl.style.display =\n",
              "          google.colab.kernel.accessAllowed ? 'block' : 'none';\n",
              "\n",
              "        async function convertToInteractive(key) {\n",
              "          const element = document.querySelector('#df-cf7ee0da-41aa-44e0-b3c0-5d84a8bfb073');\n",
              "          const dataTable =\n",
              "            await google.colab.kernel.invokeFunction('convertToInteractive',\n",
              "                                                     [key], {});\n",
              "          if (!dataTable) return;\n",
              "\n",
              "          const docLinkHtml = 'Like what you see? Visit the ' +\n",
              "            '<a target=\"_blank\" href=https://colab.research.google.com/notebooks/data_table.ipynb>data table notebook</a>'\n",
              "            + ' to learn more about interactive tables.';\n",
              "          element.innerHTML = '';\n",
              "          dataTable['output_type'] = 'display_data';\n",
              "          await google.colab.output.renderOutput(dataTable, element);\n",
              "          const docLink = document.createElement('div');\n",
              "          docLink.innerHTML = docLinkHtml;\n",
              "          element.appendChild(docLink);\n",
              "        }\n",
              "      </script>\n",
              "    </div>\n",
              "  </div>\n",
              "  "
            ]
          },
          "metadata": {},
          "execution_count": 14
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "As we can see, these are mostly not \"normal\" words. Let's take an example of \"strongertogether\" which occurs as much as 196 in expert tweets and only once in a 'not expert' tweet. It is a popular hashtag to support Ukraine. Sometimes it is written without a # sign, so it was not filtered and remained in the tweets body.\n",
        "\n",
        "This tells us a lot about the nature of this project. Some words, strongly incriminating expert/non-expert status, may be of a temporal nature. Popular hash tags may disappear after a few weeks and their value for a model may diminish. The other may come in their place. It's something different to kaggle contests and we must remember about this in real world project."
      ],
      "metadata": {
        "id": "8cgRaE-VzbgF"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "freq[freq['total'] > 10].sort_values('not_expert_rate', ascending = False).head(20)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 669
        },
        "id": "SohIXqRjTv5x",
        "outputId": "91a9d71d-cfbe-47b7-c9d9-5512d42bdab9"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "                not  expert  total  expert_rate  not_expert_rate\n",
              "peoples          11       0     11     0.000000         1.000000\n",
              "ukrainianarmy    22       0     22     0.000000         1.000000\n",
              "ruzzia           16       1     17     0.058824         0.941176\n",
              "thats            15       1     16     0.062500         0.937500\n",
              "dude             15       1     16     0.062500         0.937500\n",
              "translation      13       1     14     0.071429         0.928571\n",
              "lmao             13       1     14     0.071429         0.928571\n",
              "im               13       1     14     0.071429         0.928571\n",
              "ukrainewarnews   12       1     13     0.076923         0.923077\n",
              "ur               11       1     12     0.083333         0.916667\n",
              "rus              11       1     12     0.083333         0.916667\n",
              "clown            10       1     11     0.090909         0.909091\n",
              "ya               10       1     11     0.090909         0.909091\n",
              "fine             18       2     20     0.100000         0.900000\n",
              "damn             18       2     20     0.100000         0.900000\n",
              "bless            16       2     18     0.111111         0.888889\n",
              "dont             28       4     32     0.125000         0.875000\n",
              "crap             13       2     15     0.133333         0.866667\n",
              "bc               13       2     15     0.133333         0.866667\n",
              "shame            19       3     22     0.136364         0.863636"
            ],
            "text/html": [
              "\n",
              "  <div id=\"df-3599cb0d-d841-4590-b7e3-a96665e9b36e\">\n",
              "    <div class=\"colab-df-container\">\n",
              "      <div>\n",
              "<style scoped>\n",
              "    .dataframe tbody tr th:only-of-type {\n",
              "        vertical-align: middle;\n",
              "    }\n",
              "\n",
              "    .dataframe tbody tr th {\n",
              "        vertical-align: top;\n",
              "    }\n",
              "\n",
              "    .dataframe thead th {\n",
              "        text-align: right;\n",
              "    }\n",
              "</style>\n",
              "<table border=\"1\" class=\"dataframe\">\n",
              "  <thead>\n",
              "    <tr style=\"text-align: right;\">\n",
              "      <th></th>\n",
              "      <th>not</th>\n",
              "      <th>expert</th>\n",
              "      <th>total</th>\n",
              "      <th>expert_rate</th>\n",
              "      <th>not_expert_rate</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>peoples</th>\n",
              "      <td>11</td>\n",
              "      <td>0</td>\n",
              "      <td>11</td>\n",
              "      <td>0.000000</td>\n",
              "      <td>1.000000</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>ukrainianarmy</th>\n",
              "      <td>22</td>\n",
              "      <td>0</td>\n",
              "      <td>22</td>\n",
              "      <td>0.000000</td>\n",
              "      <td>1.000000</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>ruzzia</th>\n",
              "      <td>16</td>\n",
              "      <td>1</td>\n",
              "      <td>17</td>\n",
              "      <td>0.058824</td>\n",
              "      <td>0.941176</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>thats</th>\n",
              "      <td>15</td>\n",
              "      <td>1</td>\n",
              "      <td>16</td>\n",
              "      <td>0.062500</td>\n",
              "      <td>0.937500</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>dude</th>\n",
              "      <td>15</td>\n",
              "      <td>1</td>\n",
              "      <td>16</td>\n",
              "      <td>0.062500</td>\n",
              "      <td>0.937500</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>translation</th>\n",
              "      <td>13</td>\n",
              "      <td>1</td>\n",
              "      <td>14</td>\n",
              "      <td>0.071429</td>\n",
              "      <td>0.928571</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>lmao</th>\n",
              "      <td>13</td>\n",
              "      <td>1</td>\n",
              "      <td>14</td>\n",
              "      <td>0.071429</td>\n",
              "      <td>0.928571</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>im</th>\n",
              "      <td>13</td>\n",
              "      <td>1</td>\n",
              "      <td>14</td>\n",
              "      <td>0.071429</td>\n",
              "      <td>0.928571</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>ukrainewarnews</th>\n",
              "      <td>12</td>\n",
              "      <td>1</td>\n",
              "      <td>13</td>\n",
              "      <td>0.076923</td>\n",
              "      <td>0.923077</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>ur</th>\n",
              "      <td>11</td>\n",
              "      <td>1</td>\n",
              "      <td>12</td>\n",
              "      <td>0.083333</td>\n",
              "      <td>0.916667</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>rus</th>\n",
              "      <td>11</td>\n",
              "      <td>1</td>\n",
              "      <td>12</td>\n",
              "      <td>0.083333</td>\n",
              "      <td>0.916667</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>clown</th>\n",
              "      <td>10</td>\n",
              "      <td>1</td>\n",
              "      <td>11</td>\n",
              "      <td>0.090909</td>\n",
              "      <td>0.909091</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>ya</th>\n",
              "      <td>10</td>\n",
              "      <td>1</td>\n",
              "      <td>11</td>\n",
              "      <td>0.090909</td>\n",
              "      <td>0.909091</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>fine</th>\n",
              "      <td>18</td>\n",
              "      <td>2</td>\n",
              "      <td>20</td>\n",
              "      <td>0.100000</td>\n",
              "      <td>0.900000</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>damn</th>\n",
              "      <td>18</td>\n",
              "      <td>2</td>\n",
              "      <td>20</td>\n",
              "      <td>0.100000</td>\n",
              "      <td>0.900000</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>bless</th>\n",
              "      <td>16</td>\n",
              "      <td>2</td>\n",
              "      <td>18</td>\n",
              "      <td>0.111111</td>\n",
              "      <td>0.888889</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>dont</th>\n",
              "      <td>28</td>\n",
              "      <td>4</td>\n",
              "      <td>32</td>\n",
              "      <td>0.125000</td>\n",
              "      <td>0.875000</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>crap</th>\n",
              "      <td>13</td>\n",
              "      <td>2</td>\n",
              "      <td>15</td>\n",
              "      <td>0.133333</td>\n",
              "      <td>0.866667</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>bc</th>\n",
              "      <td>13</td>\n",
              "      <td>2</td>\n",
              "      <td>15</td>\n",
              "      <td>0.133333</td>\n",
              "      <td>0.866667</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>shame</th>\n",
              "      <td>19</td>\n",
              "      <td>3</td>\n",
              "      <td>22</td>\n",
              "      <td>0.136364</td>\n",
              "      <td>0.863636</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "</div>\n",
              "      <button class=\"colab-df-convert\" onclick=\"convertToInteractive('df-3599cb0d-d841-4590-b7e3-a96665e9b36e')\"\n",
              "              title=\"Convert this dataframe to an interactive table.\"\n",
              "              style=\"display:none;\">\n",
              "        \n",
              "  <svg xmlns=\"http://www.w3.org/2000/svg\" height=\"24px\"viewBox=\"0 0 24 24\"\n",
              "       width=\"24px\">\n",
              "    <path d=\"M0 0h24v24H0V0z\" fill=\"none\"/>\n",
              "    <path d=\"M18.56 5.44l.94 2.06.94-2.06 2.06-.94-2.06-.94-.94-2.06-.94 2.06-2.06.94zm-11 1L8.5 8.5l.94-2.06 2.06-.94-2.06-.94L8.5 2.5l-.94 2.06-2.06.94zm10 10l.94 2.06.94-2.06 2.06-.94-2.06-.94-.94-2.06-.94 2.06-2.06.94z\"/><path d=\"M17.41 7.96l-1.37-1.37c-.4-.4-.92-.59-1.43-.59-.52 0-1.04.2-1.43.59L10.3 9.45l-7.72 7.72c-.78.78-.78 2.05 0 2.83L4 21.41c.39.39.9.59 1.41.59.51 0 1.02-.2 1.41-.59l7.78-7.78 2.81-2.81c.8-.78.8-2.07 0-2.86zM5.41 20L4 18.59l7.72-7.72 1.47 1.35L5.41 20z\"/>\n",
              "  </svg>\n",
              "      </button>\n",
              "      \n",
              "  <style>\n",
              "    .colab-df-container {\n",
              "      display:flex;\n",
              "      flex-wrap:wrap;\n",
              "      gap: 12px;\n",
              "    }\n",
              "\n",
              "    .colab-df-convert {\n",
              "      background-color: #E8F0FE;\n",
              "      border: none;\n",
              "      border-radius: 50%;\n",
              "      cursor: pointer;\n",
              "      display: none;\n",
              "      fill: #1967D2;\n",
              "      height: 32px;\n",
              "      padding: 0 0 0 0;\n",
              "      width: 32px;\n",
              "    }\n",
              "\n",
              "    .colab-df-convert:hover {\n",
              "      background-color: #E2EBFA;\n",
              "      box-shadow: 0px 1px 2px rgba(60, 64, 67, 0.3), 0px 1px 3px 1px rgba(60, 64, 67, 0.15);\n",
              "      fill: #174EA6;\n",
              "    }\n",
              "\n",
              "    [theme=dark] .colab-df-convert {\n",
              "      background-color: #3B4455;\n",
              "      fill: #D2E3FC;\n",
              "    }\n",
              "\n",
              "    [theme=dark] .colab-df-convert:hover {\n",
              "      background-color: #434B5C;\n",
              "      box-shadow: 0px 1px 3px 1px rgba(0, 0, 0, 0.15);\n",
              "      filter: drop-shadow(0px 1px 2px rgba(0, 0, 0, 0.3));\n",
              "      fill: #FFFFFF;\n",
              "    }\n",
              "  </style>\n",
              "\n",
              "      <script>\n",
              "        const buttonEl =\n",
              "          document.querySelector('#df-3599cb0d-d841-4590-b7e3-a96665e9b36e button.colab-df-convert');\n",
              "        buttonEl.style.display =\n",
              "          google.colab.kernel.accessAllowed ? 'block' : 'none';\n",
              "\n",
              "        async function convertToInteractive(key) {\n",
              "          const element = document.querySelector('#df-3599cb0d-d841-4590-b7e3-a96665e9b36e');\n",
              "          const dataTable =\n",
              "            await google.colab.kernel.invokeFunction('convertToInteractive',\n",
              "                                                     [key], {});\n",
              "          if (!dataTable) return;\n",
              "\n",
              "          const docLinkHtml = 'Like what you see? Visit the ' +\n",
              "            '<a target=\"_blank\" href=https://colab.research.google.com/notebooks/data_table.ipynb>data table notebook</a>'\n",
              "            + ' to learn more about interactive tables.';\n",
              "          element.innerHTML = '';\n",
              "          dataTable['output_type'] = 'display_data';\n",
              "          await google.colab.output.renderOutput(dataTable, element);\n",
              "          const docLink = document.createElement('div');\n",
              "          docLink.innerHTML = docLinkHtml;\n",
              "          element.appendChild(docLink);\n",
              "        }\n",
              "      </script>\n",
              "    </div>\n",
              "  </div>\n",
              "  "
            ]
          },
          "metadata": {},
          "execution_count": 15
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "Now let's come back to our engineered features - at count, hash count, link count and badwords count. Let's see if they differentiate the classes.\n",
        "\n",
        "We start with tweet length.\n",
        "As some tweets are very long, for the sake of this plot we'll treat them as outliers and drop all tweets longer than 400 chars (there are less than 500 such tweets in our dataframe)."
      ],
      "metadata": {
        "id": "1l6cSu19VECI"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "df400 = df[df['len'] < 400]"
      ],
      "metadata": {
        "id": "seRfjpX4eOUw"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "source": [
        "As some users have millions of followers, we'll plot a scatter with a followersCount (we call it 'num' in our dataset) on a log scale."
      ],
      "metadata": {
        "id": "Zo00h1Rveuzs"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "import matplotlib.pyplot as plt\n",
        "import seaborn as sns"
      ],
      "metadata": {
        "id": "nSmTYZ18fkcv"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "plt.scatter(np.log(df400['num']), df400['len'], s = 8)\n",
        "plt.xlabel('followers count')\n",
        "plt.ylabel('tweet length')\n",
        "plt.show()"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 313
        },
        "id": "yc-z-8RAf5R8",
        "outputId": "845d16bc-bae8-4df9-aa4d-a25ced5a955d"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "/usr/local/lib/python3.7/dist-packages/pandas/core/arraylike.py:364: RuntimeWarning: divide by zero encountered in log\n",
            "  result = getattr(ufunc, method)(*inputs, **kwargs)\n"
          ]
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ],
            "image/png": "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\n"
          },
          "metadata": {
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "No clear pattern has emerged. We will get the same with histpolot or boxplot or measure of correlation - no clear relationship. Tweet length is not a statistically significant feature for our project."
      ],
      "metadata": {
        "id": "c0t-sUE7f9uH"
      }
    },
    {
      "cell_type": "markdown",
      "source": [
        "In a next step, we'll take a look at 'at', 'hash', 'links' and 'badwords' features. All histograms below are normalized so that expert / non-expert counts are equalized."
      ],
      "metadata": {
        "id": "zgjRDxp-g-Aq"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "def print_col_hist(COL):\n",
        "    bins = len(df[COL].value_counts())\n",
        "    sns.displot(df, x = COL, hue = \"expert\", stat = \"density\", common_norm = False, bins = bins,\n",
        "                shrink = 0.7, discrete = True, legend = False)\n",
        "    plt.xticks(np.arange(4))\n",
        "    plt.legend(labels = ['expert', 'not expert'])\n",
        "    plt.show()"
      ],
      "metadata": {
        "id": "nD-bt_E_ixWd"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "print_col_hist('at')"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 369
        },
        "id": "boHBXqlbi7q_",
        "outputId": "f45740b7-a70c-4924-b0eb-62f7ac8459ea"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 360x360 with 1 Axes>"
            ],
            "image/png": "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\n"
          },
          "metadata": {
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "We see non-experts use '@name' much more often than experts. Looks like a significant feature."
      ],
      "metadata": {
        "id": "WtryNLw8jAsb"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "print_col_hist('hash')"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 369
        },
        "id": "_4Gm508NjRIA",
        "outputId": "3ebaa264-6500-4ad7-8dd9-9ac0a3c4ce41"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 360x360 with 1 Axes>"
            ],
            "image/png": "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\n"
          },
          "metadata": {
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "There are much fewer hashes than @ in our tweets, but if they appear, we see them much more often in experts' tweets."
      ],
      "metadata": {
        "id": "i7ELKwiZjVtZ"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "print_col_hist('links')"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 369
        },
        "id": "LKkXA3RAjiiR",
        "outputId": "98cb3801-d971-43c7-ea68-20ad34981349"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 360x360 with 1 Axes>"
            ],
            "image/png": "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\n"
          },
          "metadata": {
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "Again - experts link to the websites much more often than non-experts."
      ],
      "metadata": {
        "id": "_xH-VYysjlt9"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "print_col_hist('badwords')"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 369
        },
        "id": "6LkEpRgxjxIm",
        "outputId": "57ce289c-c36d-4be6-af4e-91387ff50fa3"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 360x360 with 1 Axes>"
            ],
            "image/png": "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\n"
          },
          "metadata": {
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "Honestly, there are much fewer badwords in our tweets than I expected, but our intuition that if used, they would appear more often in non-expert tweets proves to be true."
      ],
      "metadata": {
        "id": "33-urQTfj0HQ"
      }
    },
    {
      "cell_type": "markdown",
      "source": [
        "**BUILDING A MODEL**"
      ],
      "metadata": {
        "id": "c8mc6JCblqUj"
      }
    },
    {
      "cell_type": "markdown",
      "source": [
        "It's high time to build our model.\n",
        "Before we dive deep into exploring vocabulary, let's build a base model using only our engineered features - at, links, hash and badwords count.\n",
        "\n",
        "We'll use four simple models:\n",
        "\n",
        "*   Logistic Regression\n",
        "*   Lasso\n",
        "*   Ridge\n",
        "*   Naive Bayes\n",
        "\n",
        "We'll test all of them on a 5-fold StratifiedKFold.\n"
      ],
      "metadata": {
        "id": "xZs9IVnNlv6M"
      }
    },
    {
      "cell_type": "markdown",
      "source": [
        "**Choosing metrics**\n",
        "Before we move forward with the models (wich are very basic and you'll learn nothing new here, honestly), it's time to take a look at the metrics.\n",
        "\n",
        "Our dataset is very unbalanced - just to remind - only 6,8% of our tweets come from experts."
      ],
      "metadata": {
        "id": "u8vrsnzrpXaC"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "sns.countplot(data = df, x = 'expert')"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 296
        },
        "id": "M-ADsibUpqZ1",
        "outputId": "a5fdd53e-8f42-445f-b88c-ce0a08543b48"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<matplotlib.axes._subplots.AxesSubplot at 0x7fec66750e90>"
            ]
          },
          "metadata": {},
          "execution_count": 22
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ],
            "image/png": "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\n"
          },
          "metadata": {
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "That makes accuracy a very weak and misleading metric, of course. Just by labelling each record as \"not expert\" we could reach almost 93% accuracy."
      ],
      "metadata": {
        "id": "lccHc1PvqR8o"
      }
    },
    {
      "cell_type": "markdown",
      "source": [
        "A natural choice of metrics for imbalanced datasets are precision, recall and their combination called F1.\n",
        "\n",
        "But here really interesting things begin. Let's begin with a LogisticRegression model and F1 as a metric."
      ],
      "metadata": {
        "id": "xurFEdwJrezo"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "from sklearn.linear_model import LogisticRegression, Lasso, Ridge\n",
        "from sklearn.naive_bayes import MultinomialNB\n",
        "from sklearn.model_selection import StratifiedKFold, train_test_split\n",
        "from sklearn.metrics import classification_report, confusion_matrix\n",
        "from sklearn.metrics import f1_score"
      ],
      "metadata": {
        "id": "-ZTdL3ZCr9I4"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "cols = 'at hash links badwords'.split()\n",
        "X = df[cols] # feature scaling with any Scaler not necessary for linear models nor for Naive Bayes\n",
        "y = df['expert']"
      ],
      "metadata": {
        "id": "7LaVi5h0tJRi"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "xtrain, xtest, ytrain, ytest = train_test_split(X, y, test_size = 0.2)"
      ],
      "metadata": {
        "id": "1r7ZTR8ovuSE"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "model = LogisticRegression()\n",
        "model.fit(xtrain, ytrain)\n",
        "ypreds = model.predict(xtest)"
      ],
      "metadata": {
        "id": "4BmFEMhewETE"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "source": [
        "Let's see the results. What's the distribution of classes in our ypreds array."
      ],
      "metadata": {
        "id": "-ztlnyttx-9e"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "unique, counts = np.unique(ypreds, return_counts=True)\n",
        "print(np.asarray((unique, counts)).T)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "lSo0WRLOwvAI",
        "outputId": "bc0c9617-0923-4bb1-8b42-bf0860c6005e"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "[[    0 19947]]\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "It looks like our model predicted all values as \"not expert\".\n",
        "It resulted in a confusion matrix and F1 measures like below."
      ],
      "metadata": {
        "id": "SNF0yaA_yIsm"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "print(confusion_matrix(ytest, ypreds))"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "oC88cCEPxtWO",
        "outputId": "a83b7b9a-afd4-491e-bb91-bc242d9bec10"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "[[18559     0]\n",
            " [ 1388     0]]\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "print(f1_score(ytest, ypreds))"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "X9IDvxChx4Rc",
        "outputId": "b23a648e-4775-4795-f674-1cd0eb0b3c91"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "0.0\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "The problem is in inbalanced nature of the dataset.\n",
        "Our model.predict() method uses as a default a 0.5 threshold to divide a dataset to the classes in a binary classification.\n",
        "\n",
        "There are a couple of methods to deal with this challenge. My favourite one is at the same time the simplest. In such a case I always choose brute force. A function below checks F1 metric for all thresholds in a range from 0.01 to 0.99 and picks the one with the highest metric."
      ],
      "metadata": {
        "id": "4ahqpco_zgf7"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "ypreds_proba = model.predict_proba(xtest)"
      ],
      "metadata": {
        "id": "LRjQ6-4k1XLG"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "ypreds_temp = np.zeros(ypreds.shape[0])\n",
        "\n",
        "f1scores = []\n",
        "for t in np.arange(0.01, 1, 0.01):\n",
        "    ypreds_temp = np.where(ypreds_proba[:, 0] > t, 0, 1)\n",
        "    f1 = f1_score(ytest, ypreds_temp)\n",
        "    f1scores.append((t, f1))"
      ],
      "metadata": {
        "id": "BZgEtxHM0arB"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "source": [
        "f1scores is a list of tuples. Each tuple contains a threshold level 't' and an F1 metric for a given threashold.\n",
        "\n",
        "Let's see how it looks like on a plot."
      ],
      "metadata": {
        "id": "ggDP7QJd15ha"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "f1scores = pd.DataFrame(f1scores)\n",
        "f1scores.columns = 'thresh f1'.split()\n",
        "threshold = f1scores.loc[f1scores['f1'].idxmax(), 'thresh']\n",
        "max_f1 = f1scores['f1'].max()\n",
        "\n",
        "plt.scatter(f1scores.iloc[:, 0], f1scores.iloc[:, 1], s = 10)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 282
        },
        "id": "LqQp6in51sb7",
        "outputId": "e1c7f5e5-3701-487a-ce59-6f6dfbb07906"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<matplotlib.collections.PathCollection at 0x7fec6681a950>"
            ]
          },
          "metadata": {},
          "execution_count": 32
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ],
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXoAAAD4CAYAAADiry33AAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADh0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uMy4yLjIsIGh0dHA6Ly9tYXRwbG90bGliLm9yZy+WH4yJAAASd0lEQVR4nO3df2xdZ33H8fe3Nw7NRKGl8cLW5heqK2FgatAlpYJ0TK3bgLRG2jpIEGo7qlUwukpjmlSERFn4Y2No+yNSpZKJaDCpLRRpYGlDlQsFAmq73BDoGk9d0vRXOhpM01WTaohjf/fHvfFuPCc+TnzvtR+/X5LVc+55zj3fp44+9/Fzzj0nMhNJUrku6HUBkqTOMuglqXAGvSQVzqCXpMIZ9JJUuBW9LmCm1atX54YNG3pdhiQtKfv37/9lZvbPtm3RBf2GDRtoNBq9LkOSlpSIeP5M25y6kaTCGfSSVDiDXpIKZ9BLUuEMekkqnEEvSYVbdJdXStJiNjJ6jL2Hxtgy0M/Q4Jpel1OJQS9p2WkPa2DW5dlCfGT0GHc9cIDxiUkeahzl4+/fyP/8amK6/WL9EIjFdj/6er2efmFKUqe0h/XKWnP2+sTk1GnLq/pqp4U4ND8AXjz+Oo8+PTb9XrULgsmpnG6/50fPMj4xyaq+Grt2bOpq2EfE/sysz7bNEb2kZWXvoTHGJyaBZqif0r48PjHJfT94hsmp5MF/e3F6+8raBaysXcCJySlqAZNTOd3+kdGXp993fGKSvYfGFs2o3pOxkpaVLQP9rOqrAUwH98zl9hA/MTk1/SFwYnKK911xKbdcs55PfOCK6fdZ1Vfj+sG3nrZ+0YV9fO7bTzEyeqyr/ZuNI3pJy8rQ4Bp27dh01jn6iy7sm56GmTml89Gr10+P1K9ae/Fpc/Kn1tv3f6hxtOvTODM5Ry9JszjTCdsqgf25bz/F1x77v3uM3XLNenZue2fHagXn6CVp3oYG15wW6vMZkW8Z6OehxtHpE7OnPix6xaCXpAU2c3qo1ydlDXpJ6oCZfxH0klfdSFLhDHpJKpxBL0mFM+glqXAGvSQVzqCXpMIZ9JJUOINekgpn0EtS4Qx6SSqcQS9JhTPoJalwBr0kFc6gl6TCGfSSVDiDXpIKVynoI2JrRDwdEYcj4u5Ztn86IkYj4smI+G5ErG/bdmtEHGr93LqQxUuS5jZn0EdEDbgX+CAwCOyIiMEZzQ4A9cz8HeCbwN+29n0LcA9wNbAZuCciLlm48iVJc6kyot8MHM7MI5l5AngQ2NbeIDMfzczXW6uPA5e3lm8ERjLzeGa+CowAWxemdElSFVWC/jLgxbb1o63XzuR24Dvz2Tci7oiIRkQ0xsbGKpQkSapqQU/GRsTHgDrwpfnsl5m7M7OemfX+/v6FLEmSlr0qQf8SsLZt/fLWa6eJiOuBzwI3Zeav57OvJKlzqgT9PmAgIjZGxEpgOzDc3iAiNgFfphnyv2jb9DBwQ0Rc0joJe0PrNUlSl6yYq0FmnoyIO2kGdA3Yk5kHI2In0MjMYZpTNW8EHooIgBcy86bMPB4RX6D5YQGwMzOPd6QnkqRZRWb2uobT1Ov1bDQavS5DkpaUiNifmfXZtvnNWEkqnEEvSYUz6CWpcAa9JBXOoJekwhn0klQ4g16SCmfQS1LhDHpJKpxBL0mFM+glqXAGvSQVzqCXpMIZ9JJUOINekgpn0EtS4Qx6SSqcQS9JhTPoJalwcz4cXJJ0fkZGj7H30BhbBvoZGlzT9eM7opekDhoZPcZdDxzga489z10PHGBk9FjXa3BEL0kdtPfQGOMTkwCMT0xy/xPPd31074hekjpoy0A/q/pqAKysXcCPD7/S9dG9QS9JHTQ0uIZdOzZxyzXred8Vl3Jicgpoju73HhrrSg1O3UhShw0NrmFocA0jo8d4/MhxxicmWdVXY8tAf1eOb9BLUpecGt13e47eoJekLjo1uu8m5+glqXAGvSQVzqCXpMIZ9JJUOE/GSloWen2/mV5yRC+peIvhfjO9VCnoI2JrRDwdEYcj4u5Ztl8bET+JiJMRcfOMbZMR8dPWz/BCFS5JVc2830y3vpG6WMwZ9BFRA+4FPggMAjsiYnBGsxeA24D7Z3mL8cy8qvVz03nWK0nz1n6/mW5+I3WxqDJHvxk4nJlHACLiQWAbMHqqQWY+19o21YEaJem89OobqYtFlaC/DHixbf0ocPU8jnFhRDSAk8DfZOa35rGvJC2IXnwjdbHoxlU36zPzpYh4G/C9iPj3zHymvUFE3AHcAbBu3boulCRJvdetK4GqnIx9CVjbtn5567VKMvOl1n+PAN8HNs3SZndm1jOz3t+/vObOJC1P3bwSqErQ7wMGImJjRKwEtgOVrp6JiEsi4g2t5dXA+2ib25ek5aqbVwLNGfSZeRK4E3gY+A/gG5l5MCJ2RsRNABHxnog4CvwR8OWIONja/e1AIyJ+BjxKc47eoJe07HXzSqDIzI69+bmo1+vZaDR6XYYkddxCztFHxP7MrM+2zVsgSFKPdOtKIG+BIEmFM+glqXAGvSQVzqCXpMIZ9JJUOINekgpn0EtS4Qx6SSqcX5iSpEWgk3eydEQvST3W6TtZGvSS1GOdvpOlQS9JPdbpO1k6Ry9JPdbpZ9oa9JK0CHTyTpZO3UhS4Qx6SSqcQS9JhTPoJalwBr0kFc6gl6TCGfSSVDiDXpIKZ9BLUuEMekkqnEEvSYUz6CWpcAa9JBXOoJekwhn0klQ4g16SCmfQS1LhDHpJKpxBL0mFqxT0EbE1Ip6OiMMRcfcs26+NiJ9ExMmIuHnGtlsj4lDr59aFKlySVM2cQR8RNeBe4IPAILAjIgZnNHsBuA24f8a+bwHuAa4GNgP3RMQl51+2JKmqKiP6zcDhzDySmSeAB4Ft7Q0y87nMfBKYmrHvjcBIZh7PzFeBEWDrAtQtSaqoStBfBrzYtn609VoVlfaNiDsiohERjbGxsYpvLUmnGxk9xue+/RQjo8dmXV+uVvS6AIDM3A3sBqjX69njciQtQSOjx7jrgQOMT0zyUOMoH3//Rvb86Nnp9V07NjE0uKbXZfZElRH9S8DatvXLW69VcT77SlJlew+NMT4xCcD4xCSPjL582vreQ8t3tqBK0O8DBiJiY0SsBLYDwxXf/2Hghoi4pHUS9obWa5K0oLYM9LOqrwbAqr4a1w++9bT1LQP9vSyvp+acusnMkxFxJ82ArgF7MvNgROwEGpk5HBHvAf4ZuAT4/Yj4q8x8R2Yej4gv0PywANiZmcc71BdJS9TI6DH2HhqbDuNzXd61Y9P08tDgGq5ae/Fp68tVZC6uKfF6vZ6NRqPXZUjqkva59ZW15iTDicmpeS+v6qst63n4iNifmfXZtvnNWEk91T63fmJyihOTU+e0vNzn4c/GoJfUU+1z6ytrF0yP0ue7vNzn4c9mUVxeKWn5Ghpcc9rcOpz7HP1ynbaZi3P0klQA5+glaRkz6CWpcAa9JBXOoJekwhn0klQ4g16SCmfQS1LhDHpJKpxBL0mFM+glqXAGvSQVzqCXpMIZ9JJUOINekgpn0EtS4Qx6SSqcQS9JhTPoJalwBr0kFc6gl6TCGfSSVDiDXpIKZ9BLUuEMekkqnEEvSYUz6CWpcAa9JBXOoJekwhn0klQ4g16SClcp6CNia0Q8HRGHI+LuWba/ISK+3tr+RERsaL2+ISLGI+KnrZ/7FrZ8SdJcVszVICJqwL3AEHAU2BcRw5k52tbsduDVzLwiIrYDXwQ+0tr2TGZetcB1S5IqqjKi3wwczswjmXkCeBDYNqPNNuCrreVvAtdFRCxcmZKkc1Ul6C8DXmxbP9p6bdY2mXkSeA24tLVtY0QciIgfRMSW2Q4QEXdERCMiGmNjY/PqgCTp7Dp9MvbnwLrM3AR8Grg/It40s1Fm7s7MembW+/v7O1ySJC0vVYL+JWBt2/rlrddmbRMRK4A3A69k5q8z8xWAzNwPPANceb5FS5KqqxL0+4CBiNgYESuB7cDwjDbDwK2t5ZuB72VmRkR/62QuEfE2YAA4sjClS5KqmPOqm8w8GRF3Ag8DNWBPZh6MiJ1AIzOHga8A/xQRh4HjND8MAK4FdkbEBDAFfCIzj3eiI5Kk2UVm9rqG09Tr9Ww0Gr0uQ5KWlIjYn5n12bb5zVhJKpxBL0mFM+glqXAGvSQVzqCXpMIZ9JJUOINekgpn0EtS4Qx6SSqcQS9JhTPoJalwBr0kFc6gl6TCGfSSVDiDXpIKZ9BLUuEMekkq3JyPEpSkThgZPcbeQ2NsGehnaHBNr8spmiN6SV03MnqMux44wNcee567HjjAyOixXpdUNINeUtftPTTG+MQkAOMTk+w9NNbjispm0Evqui0D/azqqwGwqq/GloH+HldUNufoJXXd0OAadu3Y5Bx9lxj0knpiaHCNAd8lTt1IUuEMekkqnEEvSYUz6CWpcAa9JBXOoJekwhn0klQ4g16SCmfQS1LhDHpJKpxBL0mFqxT0EbE1Ip6OiMMRcfcs298QEV9vbX8iIja0bftM6/WnI+LGhStdklTFnDc1i4gacC8wBBwF9kXEcGaOtjW7HXg1M6+IiO3AF4GPRMQgsB14B/DbwCMRcWVmTi50R+D0J9YAC748NLim48dYLMe2r2UeezH11RuadU9k5tkbRFwDfD4zb2ytfwYgM/+6rc3DrTaPRcQK4GWgH7i7vW17uzMdr16vZ6PRmHdHTj2xZnxikpW15h8qJyanFmx5VV+Nj79/I3t+9GzHjrFYjm1f7Wunj72qr8auHZsM+wUUEfszsz7btiq3Kb4MeLFt/Shw9ZnaZObJiHgNuLT1+uMz9r1slgLvAO4AWLduXYWS/r/2J9acmJyafn2hlscnJnlk9OWOHmOxHNu+2tdOH/vUU6UM+u5YFCdjM3N3ZtYzs97ff25Pmml/Ys3K2gXTo4mFWl7VV+P6wbd29BiL5dj21b52+tg+Vaq7qozoXwLWtq1f3npttjZHW1M3bwZeqbjvgpj5xBqgI3OMV629uGfzm908tn0t89iLqa+O5runyhz9CuA/getohvQ+4KOZebCtzaeAd2XmJ1onY/8gMz8cEe8A7gc20zwZ+11g4GwnY891jl6SlrPzmqNvzbnfCTwM1IA9mXkwInYCjcwcBr4C/FNEHAaO07zShla7bwCjwEngU5264kaSNLs5R/Td5ohekubvbCP6RXEyVpLUOQa9JBXOoJekwhn0klS4RXcyNiLGgOfnudtq4JcdKGexs9/Li/1eXubb7/WZOeu30BZd0J+LiGic6Wxzyez38mK/l5eF7LdTN5JUOINekgpXStDv7nUBPWK/lxf7vbwsWL+LmKOXJJ1ZKSN6SdIZGPSSVLglE/Tn84DypaxCvz8dEaMR8WREfDci1veizk6Yq+9t7f4wIjIiirgEr0q/I+LDrd/7wYi4v9s1dkKFf+vrIuLRiDjQ+vf+oV7UuZAiYk9E/CIinjrD9oiIXa3/J09GxLvP6UCZueh/aN4e+RngbcBK4GfA4Iw2fwrc11reDny913V3qd+/B/xGa/mTJfS7at9b7S4CfkjzkZX1Xtfdpd/5AHAAuKS1/pu9rrtL/d4NfLK1PAg81+u6F6Df1wLvBp46w/YPAd8BAngv8MS5HGepjOg3A4cz80hmngAeBLbNaLMN+Gpr+ZvAdRERXayxE+bsd2Y+mpmvt1Yfp/kUrxJU+Z0DfAH4IvCrbhbXQVX6/SfAvZn5KkBm/qLLNXZClX4n8KbW8puB/+pifR2RmT+k+QyPM9kGfC2bHgcujojfmu9xlkrQz/aA8pkPGT/tAeXAqQeUL2VV+t3udpqf/iWYs++tP2PXZua/dLOwDqvyO78SuDIifhwRj0fE1q5V1zlV+v154GMRcRT4V+DPulNaT803A2ZV5ZmxWgIi4mNAHfjdXtfSDRFxAfD3wG09LqUXVtCcvvkAzb/gfhgR78rM/+5pVZ23A/jHzPy7iLiG5lPt3pmZU70ubLFbKiP6+Tyg/NRzbk89oHwpq/Rw9Yi4HvgscFNm/rpLtXXaXH2/CHgn8P2IeI7m/OVwASdkq/zOjwLDmTmRmc/SfKbzQJfq65Qq/b4d+AZAZj4GXEjzxl8lq5QBc1kqQb8PGIiIjRGxkubJ1uEZbYaBW1vLNwPfy9bZjCVszn5HxCbgyzRDvoS52lPO2vfMfC0zV2fmhszcQPP8xE2ZudSfQ1nl3/q3aI7miYjVNKdyjnSzyA6o0u8XgOsAIuLtNIN+rKtVdt8wcEvr6pv3Aq9l5s/n+yZLYuomz+MB5UtZxX5/CXgj8FDr3PMLmXlTz4peIBX7XpyK/X4YuCEiRoFJ4C8zc0n/9Vqx338B/ENE/DnNE7O3LfXBXEQ8QPNDe3Xr3MM9QB9AZt5H81zEh4DDwOvAH5/TcZb4/ydJ0hyWytSNJOkcGfSSVDiDXpIKZ9BLUuEMekkqnEEvSYUz6CWpcP8LXxRixZakCzMAAAAASUVORK5CYII=\n"
          },
          "metadata": {
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "print(f\"Best F1 equals {max_f1:.3f} at a threshold {threshold}\")"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "ExirwA3d2L9m",
        "outputId": "a3c854b3-f851-4a0f-8c5e-8dcced7334ad"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Best F1 equals 0.231 at a threshold 0.92\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "That's interesting - we reached F1 0.23 at a threshold 0.92!\n",
        "That is far from good yet, but it's so much more than 0.00 at a 0.5 threshold :-)\n",
        "We can see on a plot any positive F1 values begin around 0.75 threshold. We'll observe the same with all other models we'll test in this notebook."
      ],
      "metadata": {
        "id": "OjWQu8552dng"
      }
    },
    {
      "cell_type": "markdown",
      "source": [
        "Let's try it on a couple of models with a StratifiedKFold"
      ],
      "metadata": {
        "id": "LM4urIha1_Pe"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "from sklearn.linear_model import LogisticRegression\n",
        "from sklearn.naive_bayes import MultinomialNB, BernoulliNB\n",
        "from sklearn.tree import DecisionTreeClassifier\n",
        "from sklearn.model_selection import StratifiedKFold\n",
        "from sklearn.metrics import f1_score"
      ],
      "metadata": {
        "id": "M6T7lYOd2Gv5"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "X = X.to_numpy()\n",
        "y = y.to_numpy()"
      ],
      "metadata": {
        "id": "mreUvQdF2ReY"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "models = [\n",
        "    LogisticRegression(),\n",
        "    DecisionTreeClassifier(),\n",
        "    BernoulliNB(),\n",
        "    MultinomialNB(),\n",
        "]\n",
        "\n",
        "all_models = []\n",
        "for model in models:\n",
        "    model_result = {}\n",
        "    fold = StratifiedKFold(n_splits = 5, shuffle = True, random_state = 100)\n",
        "    best_f1 = []\n",
        "\n",
        "    for i, (trainidx, testidx) in enumerate(fold.split(X, y)):\n",
        "        xtrain, ytrain = X[trainidx], y[trainidx]\n",
        "        xtest, ytest = X[testidx], y[testidx]\n",
        "        f1scores = []\n",
        "\n",
        "        model = model\n",
        "        model.fit(xtrain, ytrain)\n",
        "        ypreds_proba = model.predict_proba(xtest)\n",
        "        ypreds_temp = np.zeros(ypreds_proba.shape[0])\n",
        "\n",
        "        for t in np.arange(0.01, 1, 0.01):\n",
        "            ypreds_temp = np.where(ypreds_proba[:, 0] > t, 0, 1)\n",
        "            f1 = f1_score(ytest, ypreds_temp)\n",
        "            f1scores.append((t, f1))\n",
        "\n",
        "\n",
        "        f1scores_df = pd.DataFrame(f1scores)\n",
        "        f1scores_df.columns = 'thresh f1'.split()\n",
        "        threshold = f1scores_df.loc[f1scores_df['f1'].idxmax(), 'thresh']\n",
        "        max_f1 = f1scores_df['f1'].max()\n",
        "        best_f1.append(max_f1)\n",
        "\n",
        "\n",
        "    model_result['model'] = model\n",
        "    model_result['bestf1'] = np.mean(best_f1)\n",
        "    all_models.append(model_result)\n",
        "    print(f\"Best F1 result for {model} is {np.mean(best_f1):.3f}\")\n",
        "\n",
        "\n",
        "all_models_df = pd.DataFrame(all_models)\n",
        "print('\\n')\n",
        "print((all_models_df.set_index('model')))"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "jmgTav-n2d-1",
        "outputId": "f448c4c0-196e-4252-fbde-5b13cb9f5627"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Best F1 result for LogisticRegression() is 0.228\n",
            "Best F1 result for DecisionTreeClassifier() is 0.233\n",
            "Best F1 result for BernoulliNB() is 0.227\n",
            "Best F1 result for MultinomialNB() is 0.225\n",
            "\n",
            "\n",
            "                            bestf1\n",
            "model                             \n",
            "LogisticRegression()      0.228308\n",
            "DecisionTreeClassifier()  0.233300\n",
            "BernoulliNB()             0.227033\n",
            "MultinomialNB()           0.225409\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "All models produce roughtly the same best F1 result from the engineered features. I doubt more can be squeezed out of these. Let's move forward to the tweets' vocabulary!"
      ],
      "metadata": {
        "id": "Yz2Lt-RX3Swc"
      }
    },
    {
      "cell_type": "markdown",
      "source": [
        "**TFIDF**\n",
        "\n",
        "We will explore three variants:\n",
        "\n",
        "*   TFIDF with ngrams (1,1) to (1, 3) on a LogisticRegression and NaiveBayes models\n",
        "*   TFIDF with grams (1,1) to (1, 3) chi2 selection on a LogisticRegresion and NaiveBayes model\n",
        "*   TFIDF with chi2 selection and engineered features on a Keras model\n",
        "\n",
        "Our tests showed removing stopwords has a... negative impact on a result, so we'll keep stopwords in a vectorized vocabulary.\n",
        "\n",
        "I had to reduce max_features to 2000 due to memory issues.\n",
        "With max_features = 5000 F1 values would be around 0.01-0.02 higher.\n",
        "\n",
        "For simplicity we dropped kfold verification and only applied train/test split once."
      ],
      "metadata": {
        "id": "CPIu29Hl4lJv"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "from sklearn.feature_extraction.text import TfidfVectorizer\n",
        "from sklearn.model_selection import train_test_split"
      ],
      "metadata": {
        "id": "3XhJ29QkHclD"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "y = df['expert']\n",
        "models = [\n",
        "    LogisticRegression(),\n",
        "    MultinomialNB(),\n",
        "    BernoulliNB()\n",
        "]\n",
        "\n",
        "all_results = []\n",
        "for model in models:\n",
        "    for n in range(1, 4):\n",
        "        vectorizer = TfidfVectorizer(max_features = 2000, ngram_range = (1, n))\n",
        "        X = vectorizer.fit_transform(df['processed'])\n",
        "        X = X.toarray()\n",
        "        results_dict = {}\n",
        "\n",
        "        xtrain, xtest, ytrain, ytest = train_test_split(X, y, train_size = 0.8)\n",
        "        model.fit(xtrain, ytrain)\n",
        "        ypreds = model.predict(xtest)\n",
        "        # print(classification_report(ytest, ypreds))\n",
        "\n",
        "        # f1_score(ytest, ypreds)\n",
        "\n",
        "        ypreds_proba = model.predict_proba(xtest)\n",
        "\n",
        "        # unique, counts = np.unique(ypreds, return_counts=True)\n",
        "        # print(np.asarray((unique, counts)).T)\n",
        "\n",
        "        ypreds_temp = np.zeros(ypreds.shape[0])\n",
        "\n",
        "        f1scores = []\n",
        "        for t in np.arange(0, 1, 0.01):\n",
        "            ypreds_temp = np.where(ypreds_proba[:, 0] > t, 0, 1)\n",
        "            f1 = f1_score(ytest, ypreds_temp)\n",
        "            f1scores.append((t, f1))\n",
        "\n",
        "        f1scores = pd.DataFrame(f1scores)\n",
        "        f1scores.columns = 'thresh f1'.split()\n",
        "        threshold = f1scores.loc[f1scores['f1'].idxmax(), 'thresh'] # threshold przy którym f1 ma max\n",
        "        max_f1 = f1scores['f1'].max()\n",
        "\n",
        "        # plt.scatter(f1scores.iloc[:, 0], f1scores.iloc[:, 1])\n",
        "        # plt.show()\n",
        "        print(f\"Best F1 score for model {model} for ngram 1-{n} is {max_f1:.3f}\")\n",
        "        results_dict['model'] = model\n",
        "        results_dict['ngram'] = '1, ' + str(n)\n",
        "        results_dict['best_thresh'] = threshold\n",
        "        results_dict['f1'] = max_f1\n",
        "        all_results.append(results_dict)\n",
        "\n",
        "print('\\n')\n",
        "all_models_df = pd.DataFrame(all_results)\n",
        "print(all_models_df)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "Npp25ZxtHStY",
        "outputId": "abb7fb4e-f47b-44ac-e336-0f4b2d0706db"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Best F1 score for model LogisticRegression() for ngram 1-1 is 0.267\n",
            "Best F1 score for model LogisticRegression() for ngram 1-2 is 0.267\n",
            "Best F1 score for model LogisticRegression() for ngram 1-3 is 0.262\n",
            "Best F1 score for model MultinomialNB() for ngram 1-1 is 0.273\n",
            "Best F1 score for model MultinomialNB() for ngram 1-2 is 0.245\n",
            "Best F1 score for model MultinomialNB() for ngram 1-3 is 0.270\n",
            "Best F1 score for model BernoulliNB() for ngram 1-1 is 0.274\n",
            "Best F1 score for model BernoulliNB() for ngram 1-2 is 0.263\n",
            "Best F1 score for model BernoulliNB() for ngram 1-3 is 0.269\n",
            "\n",
            "\n",
            "                  model ngram  best_thresh        f1\n",
            "0  LogisticRegression()  1, 1         0.84  0.267481\n",
            "1  LogisticRegression()  1, 2         0.85  0.267408\n",
            "2  LogisticRegression()  1, 3         0.84  0.261533\n",
            "3       MultinomialNB()  1, 1         0.87  0.273246\n",
            "4       MultinomialNB()  1, 2         0.88  0.244918\n",
            "5       MultinomialNB()  1, 3         0.88  0.269793\n",
            "6         BernoulliNB()  1, 1         0.68  0.273795\n",
            "7         BernoulliNB()  1, 2         0.61  0.262731\n",
            "8         BernoulliNB()  1, 3         0.55  0.268587\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "With TFIDF, differences between models are rather insignificant and it's possible that with another train/test split the results would be slightly differnt. Bigrams and trigrams do not add value."
      ],
      "metadata": {
        "id": "zBrJ9OLe8PqW"
      }
    },
    {
      "cell_type": "markdown",
      "source": [
        "**TFIDF wich chi2 vocabulary word selection**\n",
        "Let's select top 1000 words with the strongest association with the labelled classes.\n",
        "Again - we will use rather train_test_split than kfold for the speed."
      ],
      "metadata": {
        "id": "RnGrdugF8fGz"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "import pandas as pd\n",
        "import numpy as np\n",
        "from sklearn.feature_extraction.text import TfidfVectorizer\n",
        "from sklearn.linear_model import LogisticRegression\n",
        "from sklearn.model_selection import train_test_split\n",
        "from sklearn.metrics import classification_report, confusion_matrix\n",
        "from sklearn.metrics import f1_score, precision_score, recall_score\n",
        "import matplotlib.pyplot as plt\n",
        "from sklearn.feature_selection import chi2"
      ],
      "metadata": {
        "id": "yTGQm0m8-EYk"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "df = pd.read_csv('/content/drive/MyDrive/Python_data/processed_tweets.csv')\n",
        "df.dropna(inplace = True)\n",
        "df.reset_index(inplace = True)\n",
        "print(df.shape)\n",
        "y = df['expert']"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "PlPwO4Ho-gGz",
        "outputId": "502b8ae2-122c-440f-bae5-9f37725a55fe"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "(99731, 11)\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "temp_vectorizer = TfidfVectorizer()\n",
        "X_ = temp_vectorizer.fit_transform(df['processed'])\n",
        "n_features=1000 # we'll keep 1000 best words in our vocabulary\n",
        "chi2score = chi2(X_,y)[0]\n",
        "wscores = zip(temp_vectorizer.get_feature_names_out(), chi2score)\n",
        "wchi2 = sorted(wscores, key = lambda x:x[1])\n",
        "topchi2 = wchi2[-n_features:]\n",
        "labels = [t for t, ch in topchi2]\n",
        "vectorizer = TfidfVectorizer(vocabulary=labels)\n",
        "X = vectorizer.fit_transform(df['processed'])"
      ],
      "metadata": {
        "id": "V1dAS0Ct-nu2"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "source": [
        "For the future use in a Keras model, we'll keep a file with top 5000 words, not top 1000 words. We limit the vocab to 1000 words here as with a sparce 5000 column matrix, our model explodes. With Keras, we'll apply a dense embedding matrix, so we can accept 5000 top words."
      ],
      "metadata": {
        "id": "iM79d1tSLkSp"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "textfile = open(\"/content/drive/MyDrive/Python_data/vocabulary_chi2_5000.txt\", \"w\")\n",
        "for element in labels:\n",
        "    textfile.write(element + \"\\n\")\n",
        "textfile.close()"
      ],
      "metadata": {
        "id": "iTl6S5poIe_M"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "source": [
        "Let's plot 20 strongest words, with the highest chi2 score."
      ],
      "metadata": {
        "id": "QqGaEhoJAqNN"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "topchi2_plot = list(zip(*wchi2[-20:]))\n",
        "x = range(len(topchi2_plot[1]))\n",
        "labels = topchi2_plot[0]\n",
        "plt.figure(figsize=(12,6))\n",
        "plt.barh(x,topchi2_plot[1], align='center', alpha=0.2)\n",
        "plt.plot(topchi2_plot[1], x, '-o', markersize=5, alpha=0.8)\n",
        "plt.yticks(x, labels)\n",
        "plt.xlabel('$\\chi^2$')"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 411
        },
        "id": "etyya8EwAhbA",
        "outputId": "61b70c86-b723-45f4-82c7-a51f8f5ce362"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "Text(0.5, 0, '$\\\\chi^2$')"
            ]
          },
          "metadata": {},
          "execution_count": 6
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 864x432 with 1 Axes>"
            ],
            "image/png": "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\n"
          },
          "metadata": {
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "As we can see, this list is very similar to the list generated with our simple \"exptert/non-expert rate\" metric produced beforehead."
      ],
      "metadata": {
        "id": "t-x719xKAwGO"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "xtrain, xtest, ytrain, ytest = train_test_split(X, y, train_size = 0.8, shuffle = True, stratify = y)"
      ],
      "metadata": {
        "id": "8f3aJkpUCyya"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "model = LogisticRegression(max_iter = 200)\n",
        "model.fit(xtrain, ytrain)\n",
        "ypreds = model.predict(xtest)\n",
        "print(classification_report(ytest, ypreds))\n",
        "print('\\n')\n",
        "print(confusion_matrix(ytest, ypreds))"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "6ZAyGJUIC0oQ",
        "outputId": "0b49e6da-5d97-4955-dfa3-05c8f103a7e7"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "              precision    recall  f1-score   support\n",
            "\n",
            "           0       0.94      1.00      0.97     18586\n",
            "           1       0.93      0.06      0.12      1361\n",
            "\n",
            "    accuracy                           0.94     19947\n",
            "   macro avg       0.94      0.53      0.54     19947\n",
            "weighted avg       0.94      0.94      0.91     19947\n",
            "\n",
            "\n",
            "\n",
            "[[18580     6]\n",
            " [ 1275    86]]\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "With a standard 0.5 threshold, we received F1 score for \"expert\" guesses. We have a very nice precision - 93% of our guesses \"it's an expert\" proved to be true. However, recall tells we recognized properly only 6% experts. Together, they sum up to a poor 0.12 F1 score.\n",
        "\n",
        "Let's see how it performs when we search for the optimal threshold."
      ],
      "metadata": {
        "id": "dzZQEXr9DnOs"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "ypreds_proba = model.predict_proba(xtest)\n",
        "ypreds_temp = np.zeros(ypreds.shape[0])\n",
        "\n",
        "f1scores = []\n",
        "for t in np.arange(0, 1, 0.01):\n",
        "    ypreds_temp = np.where(ypreds_proba[:, 0] > t, 0, 1)\n",
        "    f1 = f1_score(ytest, ypreds_temp, zero_division = 0)\n",
        "    prec = precision_score(ytest, ypreds_temp, zero_division = 0)\n",
        "    rec = recall_score(ytest, ypreds_temp, zero_division = 0)\n",
        "    f1scores.append((t, f1, prec, rec))\n",
        "\n",
        "f1scores = pd.DataFrame(f1scores)\n",
        "f1scores.columns = 'thresh f1 precision recall'.split()\n",
        "threshold = f1scores.loc[f1scores['f1'].idxmax(), 'thresh'] \n",
        "max_prec = f1scores.loc[f1scores['f1'].idxmax(), 'precision'] \n",
        "max_rec = f1scores.loc[f1scores['f1'].idxmax(), 'recall'] \n",
        "max_f1 = f1scores.loc[f1scores['f1'].idxmax(), 'f1']\n",
        "\n",
        "\n",
        "print(f\"Best F1 {max_f1:.3f} at threshold {threshold} with precision {max_prec:.3f} and recall {max_rec:.3f}\")"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "BXVGxVKdEuxW",
        "outputId": "623b8b06-3482-41f6-eaab-a48d360ef75a"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Best F1 0.305 at threshold 0.87 with precision 0.251 and recall 0.388\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "Again, with an optimal threshold, we get a significantly better F1 score than with a standard 0.5 threshold. Precision and recall are both far from perfect, but at least they are more balanced now.\n",
        "\n",
        "Let's try the same with a Naive Bayes model."
      ],
      "metadata": {
        "id": "gaBwrVFrJcG-"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "from sklearn.naive_bayes import BernoulliNB"
      ],
      "metadata": {
        "id": "1nQyUoV4MKYb"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "model = BernoulliNB()\n",
        "model.fit(xtrain, ytrain)\n",
        "ypreds = model.predict(xtest)\n",
        "print(classification_report(ytest, ypreds))\n",
        "print('\\n')\n",
        "print(confusion_matrix(ytest, ypreds))"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "yAVDJdzsMQnr",
        "outputId": "800c3e9e-fa2e-4ee8-9bd4-6e2d3fce5da8"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "              precision    recall  f1-score   support\n",
            "\n",
            "           0       0.95      0.96      0.95     18586\n",
            "           1       0.35      0.29      0.31      1361\n",
            "\n",
            "    accuracy                           0.92     19947\n",
            "   macro avg       0.65      0.62      0.63     19947\n",
            "weighted avg       0.91      0.92      0.91     19947\n",
            "\n",
            "\n",
            "\n",
            "[[17866   720]\n",
            " [  972   389]]\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "It looks much more promising than Logistic Regression!\n",
        "Let's search for the optimal threshold."
      ],
      "metadata": {
        "id": "lWWFDlY7MYEJ"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "ypreds_proba = model.predict_proba(xtest)\n",
        "ypreds_temp = np.zeros(ypreds.shape[0])\n",
        "\n",
        "f1scores = []\n",
        "for t in np.arange(0, 1, 0.01):\n",
        "    ypreds_temp = np.where(ypreds_proba[:, 0] > t, 0, 1)\n",
        "    f1 = f1_score(ytest, ypreds_temp, zero_division = 0)\n",
        "    prec = precision_score(ytest, ypreds_temp, zero_division = 0)\n",
        "    rec = recall_score(ytest, ypreds_temp, zero_division = 0)\n",
        "    f1scores.append((t, f1, prec, rec))\n",
        "\n",
        "f1scores = pd.DataFrame(f1scores)\n",
        "f1scores.columns = 'thresh f1 precision recall'.split()\n",
        "threshold = f1scores.loc[f1scores['f1'].idxmax(), 'thresh'] \n",
        "max_prec = f1scores.loc[f1scores['f1'].idxmax(), 'precision'] \n",
        "max_rec = f1scores.loc[f1scores['f1'].idxmax(), 'recall'] \n",
        "max_f1 = f1scores.loc[f1scores['f1'].idxmax(), 'f1'] \n",
        "\n",
        "\n",
        "print(f\"Best F1 {max_f1:.3f} at threshold {threshold} with precision {max_prec:.3f} and recall {max_rec:.3f}\")"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "VbFiXB3kMdvr",
        "outputId": "52be60b6-a5b9-4a57-913c-bf2a0584d206"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Best F1 0.323 at threshold 0.58 with precision 0.334 and recall 0.313\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "Let's see a relationship between threshold and F1 on a plot."
      ],
      "metadata": {
        "id": "043f2oezNKbu"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "plt.scatter(f1scores.iloc[:, 0], f1scores.iloc[:, 1], s = 10, label = 'F1')\n",
        "plt.scatter(f1scores.iloc[:, 0], f1scores.iloc[:, 2], s = 10, label = 'Precision')\n",
        "plt.scatter(f1scores.iloc[:, 0], f1scores.iloc[:, 3], s = 10, label = 'Recall')\n",
        "plt.legend()\n",
        "plt.show()"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 265
        },
        "id": "4GDm6zfGNPop",
        "outputId": "37ed90b8-888f-444d-c566-1ae51ad8c559"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ],
            "image/png": "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\n"
          },
          "metadata": {
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "It's quite intutive. The higher the threshold, the higher the recall (we recognize more experts out of the pool), and the lower the precision (more and more of our guesses \"it's an expert!\" fail. F1 is maximized somewhere at a point where precision and recall plots cross. Nice exercice to understand this relationship."
      ],
      "metadata": {
        "id": "bCGlVLdvNx5k"
      }
    },
    {
      "cell_type": "markdown",
      "source": [
        "**TFIDF summary**\n",
        "\n",
        "We clearly see chi2 vocabulary selection adds around 0.02-0.03 to our F1 score. Not much, but it's a stable surplus with a smaller vocab, which results in lower memory consumption. Not bad."
      ],
      "metadata": {
        "id": "1bya31SFTSSn"
      }
    },
    {
      "cell_type": "markdown",
      "source": [
        "**Deep Learning model with KERAS**\n",
        "\n",
        "Let's try something more sophisticated to check if a deep learning model can outperform basic models like LogisticRegression or Naive Bayes.\n",
        "\n",
        "Moreover, we'll merge information from two sources - vocabulary and engineered features - in one model. To do this, we'll employ Keras API to concatenate two inputs. We will also apply chi2 word selection, which means we'll put all the best that we have to this pot."
      ],
      "metadata": {
        "id": "xoicE_0rRLq9"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "import pandas as pd\n",
        "import numpy as np\n",
        "from sklearn.metrics import f1_score, precision_score, recall_score\n",
        "from keras.preprocessing.text import Tokenizer\n",
        "from tensorflow.keras.preprocessing.sequence import pad_sequences\n",
        "from sklearn.preprocessing import MinMaxScaler\n",
        "from keras.models import Model\n",
        "from keras.layers import Flatten, Embedding, Dense, Convolution1D, MaxPooling1D, Dropout, Input, Normalization\n",
        "from keras.layers.merging import concatenate\n",
        "import tensorflow as tf\n",
        "from keras.utils.vis_utils import plot_model\n",
        "from keras.callbacks import LearningRateScheduler\n",
        "from sklearn.model_selection import StratifiedKFold\n",
        "import matplotlib.pyplot as plt\n",
        "import os"
      ],
      "metadata": {
        "id": "-fcg_J2oTHMO"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "df = pd.read_csv('/content/drive/MyDrive/Python_data/processed_tweets.csv')\n",
        "df.dropna(inplace = True)\n",
        "df.reset_index(inplace = True)\n",
        "print(df.shape)\n",
        "y = df['expert']\n",
        "X = df['processed']"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "wp_JJ3qMIAgP",
        "outputId": "91fc9cfe-dfb5-4225-cbb9-8b5fb1227fe4"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "(99731, 11)\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "tf.config.run_functions_eagerly(True)\n",
        "tf.data.experimental.enable_debug_mode()\n",
        "os.environ['TF_CPP_MIN_LOG_LEVEL'] = '3'\n",
        "\n",
        "es = tf.keras.callbacks.EarlyStopping(\n",
        "    monitor='val_loss', min_delta=1e-05, patience=5, verbose=0,\n",
        "    mode='min', baseline=None, restore_best_weights=True)\n",
        "\n",
        "plateau = tf.keras.callbacks.ReduceLROnPlateau(\n",
        "    monitor='val_loss', factor=0.7, patience=3, verbose=0,\n",
        "    mode='min')\n",
        "\n",
        "def scheduler(epoch, lr):\n",
        "    if epoch < 3:\n",
        "        return lr\n",
        "    else:\n",
        "        return lr * tf.math.exp(-0.1)\n",
        "\n",
        "lrate = LearningRateScheduler(scheduler)"
      ],
      "metadata": {
        "id": "DVTPvkhsIPjH"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "with open('/content/drive/MyDrive/Python_data/vocabulary_chi2_5000.txt') as f:\n",
        "    words = f.readlines()\n",
        "chi2_words = [word.replace('\\n', '') for word in words]"
      ],
      "metadata": {
        "id": "-EKUeRkDIPNE"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "print(chi2_words[:20])"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "gOiSzzQJJHxi",
        "outputId": "1c943a33-61a8-4208-8c12-86b8736a6265"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "['howard', 'retrain', 'ministers', 'govt', 'exploded', 'delightful', 'deeper', 'average', 'backed', 'firm', 'opting', 'superb', 'respectable', 'slope', 'loosing', 'oskil', 'dodger', 'stopfake', 'intertwined', 'krivoy']\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "Now we'll remove words not in a list chi2_words from our tweets.\n",
        "In case there was a tweet with no words from a list, we'll put \"ukraine\" in this place to avoid empty rows."
      ],
      "metadata": {
        "id": "k8EntPyRJ32K"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "vocab_size = len(set(chi2_words))\n",
        "\n",
        "def choose_chi2(sentence):\n",
        "    sent = [word for word in sentence.split() if word in chi2_words]\n",
        "    if len(sent) > 0:\n",
        "        return \" \".join(sent)\n",
        "    else:\n",
        "        return \"ukraine\"\n",
        "\n",
        "X = X.apply(lambda x: choose_chi2(x))"
      ],
      "metadata": {
        "id": "PVXR8KdQJV4l"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "for i in range(10):\n",
        "  print(X[i])\n",
        "\n",
        "# I don't know why, but when simply prompting print[:10], colab cuts the sentences; surely it can be modified in the options"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "MKhRNKeWKMrn",
        "outputId": "2b80722e-075f-48f2-b84b-5ae2d0e2a97c"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "please send ukraine urgently\n",
            "if anything see potential in and sweden and in of ukraine an so if they need that ll be sure and\n",
            "this probably as if we are of from but not or going after lost you get what like ground or like ukraine\n",
            "look you too can nazis in ukraine just donate now because nazis need money too\n",
            "pandemic not ones launder billions ukraine by energy crisis lie about being by energy crisis hope no from trump\n",
            "stand with people of ukraine and russia you military of two countries we are not same\n",
            "did you see interview with fm she much said fuck we re gonna ukraine no what they or think or or that well enough they don think they have worry\n",
            "world not ukraine good job russia all peoples without of nuclear of\n",
            "why would they want him there he all taxpayer dollars ukraine pacific and he look like at\n",
            "general staff russian troops in southern forces of ukraine strike military facilities on number of strikes carried out with strike aircraft helicopters unmanned\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "X_len = [len(x.split()) for x in X]\n",
        "max_len = max(X_len)\n",
        "min_len = min(X_len)\n",
        "print(f\"Longest sentence has {max_len} words.\")\n",
        "plt.hist(X_len)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 350
        },
        "id": "u5QK-iDYKrwN",
        "outputId": "c6320b3d-f87e-4970-ccf3-aa910c317af4"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Longest sentence has 45 words.\n"
          ]
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "(array([17311., 20587., 19652., 13230., 11954., 10840.,  4279.,  1661.,\n",
              "          188.,    29.]),\n",
              " array([ 1. ,  5.4,  9.8, 14.2, 18.6, 23. , 27.4, 31.8, 36.2, 40.6, 45. ]),\n",
              " <a list of 10 Patch objects>)"
            ]
          },
          "metadata": {},
          "execution_count": 9
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ],
            "image/png": "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\n"
          },
          "metadata": {
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "tokenizer = Tokenizer(num_words=5000)\n",
        "tokenizer.fit_on_texts(X)\n",
        "sequences = tokenizer.texts_to_sequences(X)\n",
        "X = pad_sequences(sequences, maxlen = max_len)"
      ],
      "metadata": {
        "id": "j2YtrBv2Mc7N"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "print(X[:3])"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "9FN-Ple9NL9J",
        "outputId": "e75d47bb-cb98-49ff-cbf0-be0333345ba7"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "[[   0    0    0    0    0    0    0    0    0    0    0    0    0    0\n",
            "     0    0    0    0    0    0    0    0    0    0    0    0    0    0\n",
            "     0    0    0    0    0    0    0    0    0    0    0    0    0  164\n",
            "   107    1 1044]\n",
            " [   0    0    0    0    0    0    0    0    0    0    0    0    0    0\n",
            "     0    0    0    0    0    0    0    0    0    0   32  166   83  651\n",
            "     4    2  392    2    4    3    1   52   34   32   12   93   10  154\n",
            "    17  141    2]\n",
            " [   0    0    0    0    0    0    0    0    0    0    0    0    0    0\n",
            "     0    0    0    0    0    0    0    0    0   16  225   23   32   20\n",
            "    11    3   24   25    7   41   70   69  178    8   63   30   45  333\n",
            "    41   45    1]]\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "output_dim = 200\n",
        "\n",
        "def get_model():\n",
        "    inputs_tfidf = Input(shape = (max_len, ))\n",
        "    inputs_hand = Input(shape = (5, ))\n",
        "    # część obsługująca tfidf\n",
        "    x = Embedding(input_dim = vocab_size, output_dim = output_dim, input_length = max_len, mask_zero = True)(inputs_tfidf)\n",
        "    x = Convolution1D(filters = 32, kernel_size = 3, padding = 'same', activation= 'relu' , kernel_regularizer = 'l2')(x)\n",
        "    x = MaxPooling1D(pool_size = 2)(x)\n",
        "    x = Flatten()(x)\n",
        "    x = Dense(100, kernel_regularizer = 'l2', activation = 'relu')(x)\n",
        "    x = Dropout(0.4)(x)\n",
        "    x = Dense(5, activation = 'relu')(x)\n",
        "    # x = Model(inputs = inputs_tfidf, outputs = x)\n",
        "\n",
        "    # część obsługująca hand\n",
        "    y = Dense(5, activation = 'relu')(inputs_hand)\n",
        "\n",
        "    # połączenie modeli\n",
        "    # combined = concatenate([x, y], axis = 1)\n",
        "    combined = concatenate([x, y], axis = 1)\n",
        "\n",
        "    z = Dense(10, activation = 'relu')(combined)\n",
        "    z = Dropout(0.4)(z)\n",
        "\n",
        "    outputs = Dense(1, activation = 'sigmoid')(z)\n",
        "    model = Model(inputs = [inputs_tfidf, inputs_hand], outputs = outputs)\n",
        "\n",
        "    model.compile(optimizer = 'adam', loss = 'binary_crossentropy', metrics = ['acc'])\n",
        "    return model\n"
      ],
      "metadata": {
        "id": "oDqeXCyRMjqL"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "model = get_model()\n",
        "print(model.summary())\n",
        "model_path = '/content/drive/MyDrive/Python_data/model_plot.png'\n",
        "plot_model(model, to_file=model_path, show_shapes=True, show_layer_names=True)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1000
        },
        "id": "iiTfRkMiMspd",
        "outputId": "f30ef77b-fed7-404e-a522-1af7393d4bc3"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Model: \"model\"\n",
            "__________________________________________________________________________________________________\n",
            " Layer (type)                   Output Shape         Param #     Connected to                     \n",
            "==================================================================================================\n",
            " input_1 (InputLayer)           [(None, 45)]         0           []                               \n",
            "                                                                                                  \n",
            " embedding (Embedding)          (None, 45, 200)      1000000     ['input_1[0][0]']                \n",
            "                                                                                                  \n",
            " conv1d (Conv1D)                (None, 45, 32)       19232       ['embedding[0][0]']              \n",
            "                                                                                                  \n",
            " max_pooling1d (MaxPooling1D)   (None, 22, 32)       0           ['conv1d[0][0]']                 \n",
            "                                                                                                  \n",
            " flatten (Flatten)              (None, 704)          0           ['max_pooling1d[0][0]']          \n",
            "                                                                                                  \n",
            " dense (Dense)                  (None, 100)          70500       ['flatten[0][0]']                \n",
            "                                                                                                  \n",
            " dropout (Dropout)              (None, 100)          0           ['dense[0][0]']                  \n",
            "                                                                                                  \n",
            " input_2 (InputLayer)           [(None, 5)]          0           []                               \n",
            "                                                                                                  \n",
            " dense_1 (Dense)                (None, 5)            505         ['dropout[0][0]']                \n",
            "                                                                                                  \n",
            " dense_2 (Dense)                (None, 5)            30          ['input_2[0][0]']                \n",
            "                                                                                                  \n",
            " concatenate (Concatenate)      (None, 10)           0           ['dense_1[0][0]',                \n",
            "                                                                  'dense_2[0][0]']                \n",
            "                                                                                                  \n",
            " dense_3 (Dense)                (None, 10)           110         ['concatenate[0][0]']            \n",
            "                                                                                                  \n",
            " dropout_1 (Dropout)            (None, 10)           0           ['dense_3[0][0]']                \n",
            "                                                                                                  \n",
            " dense_4 (Dense)                (None, 1)            11          ['dropout_1[0][0]']              \n",
            "                                                                                                  \n",
            "==================================================================================================\n",
            "Total params: 1,090,388\n",
            "Trainable params: 1,090,388\n",
            "Non-trainable params: 0\n",
            "__________________________________________________________________________________________________\n",
            "None\n"
          ]
        },
        {
          "output_type": "execute_result",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<IPython.core.display.Image object>"
            ]
          },
          "metadata": {},
          "execution_count": 13
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "def threhsold_f1(ypreds, ytest):\n",
        "    ypreds_temp = np.zeros(ypreds.shape[0])\n",
        "\n",
        "    f1scores = []\n",
        "    for t in np.arange(0, 1, 0.01):\n",
        "        ypreds_temp = np.where(ypreds > t, 0, 1)\n",
        "        f1 = f1_score(ytest, ypreds_temp, zero_division = 0)\n",
        "        prec = precision_score(ytest, ypreds_temp, zero_division = 0)\n",
        "        rec = recall_score(ytest, ypreds_temp, zero_division = 0)\n",
        "        f1scores.append((t, f1, prec, rec))\n",
        "\n",
        "    f1scores = pd.DataFrame(f1scores)\n",
        "    f1scores.columns = 'thresh f1 precision recall'.split()\n",
        "    threshold = f1scores.loc[f1scores['f1'].idxmax(), 'thresh']\n",
        "    max_f1 = f1scores.loc[f1scores['f1'].idxmax(), 'f1']\n",
        "    max_prec = f1scores.loc[f1scores['f1'].idxmax(), 'precision']\n",
        "    max_rec = f1scores.loc[f1scores['f1'].idxmax(), 'recall']\n",
        "\n",
        "    print('\\n')\n",
        "    print(f\"The best {max_f1:.3f} at threshold {threshold} with precision {max_prec:.3f} and recall {max_rec:.3f}\")\n",
        "\n",
        "    plt.scatter(f1scores.iloc[:, 0], f1scores.iloc[:, 1], s = 6, label = 'f1')\n",
        "    plt.scatter(f1scores.iloc[:, 0], f1scores.iloc[:, 2], s = 6, label = 'precision')\n",
        "    plt.scatter(f1scores.iloc[:, 0], f1scores.iloc[:, 3], s = 6, label = 'recall')\n",
        "    plt.legend()\n",
        "    plt.show()\n",
        "    return threshold, max_f1, max_prec, max_rec"
      ],
      "metadata": {
        "id": "LKk-KWR0cx86"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "fold = StratifiedKFold(n_splits = 5, shuffle = True, random_state = 100)\n",
        "\n",
        "best_scores = []\n",
        "for n, (traini, testi) in enumerate(fold.split(X, y)):\n",
        "    scores = {}\n",
        "    # to avoid data leakage from train to test sets, we should proceed with the same procedure of chi2-selection\n",
        "    # for each subset of xtrain before applying transform to xtest\n",
        "    # but it would be very computationaly expensive for this kind of trivial experiment\n",
        "    # so I will follow an anti-leakage procedure only on x_rest data, which goes only through MinMaxScaler\n",
        "    rest_cols = 'at hash len links badwords'.split()\n",
        "    X_rest = df[rest_cols].to_numpy()\n",
        "\n",
        "    xtrain, xresttrain, ytrain = X[traini], X_rest[traini], y[traini]\n",
        "    xtest, xresttest, ytest = X[testi], X_rest[testi], y[testi]\n",
        "\n",
        "    scaler = MinMaxScaler()\n",
        "    xresttrain = scaler.fit_transform(xresttrain)\n",
        "    xresttest = scaler.transform(xresttest)\n",
        "\n",
        "    print(f\"Epoch {n+1}\")\n",
        "    model = get_model()\n",
        "    model.fit([xtrain, xresttrain], ytrain, epochs = 20, batch_size = 64, validation_split = 0.2, verbose = 0, callbacks = [es, plateau, lrate])\n",
        "\n",
        "    ypreds = model.predict([xtest, xresttest])\n",
        "    ypreds = 1 - ypreds\n",
        "    ypreds_class = np.where(ypreds > 0.5, 0, 1)\n",
        "\n",
        "    f1_score_fold = f1_score(ytest, ypreds_class)\n",
        "    threshold_best, max_f1, max_prec, max_rec = threhsold_f1(ypreds, ytest)\n",
        "    scores = {\n",
        "        'thresh_best': threshold_best,\n",
        "        'f1_05': f1_score_fold,\n",
        "        'f1_best': max_f1,\n",
        "        'precision_best': max_prec,\n",
        "        'recall_best': max_rec,\n",
        "    }\n",
        "    best_scores.append(scores)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1000
        },
        "id": "6t8FXmRuZDvs",
        "outputId": "036ba9b5-6dfd-499b-cb08-d1e24d140a7a"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Epoch 1\n",
            "624/624 [==============================] - 5s 9ms/step\n",
            "\n",
            "\n",
            "The best 0.329 at threshold 0.8 with precision 0.331 and recall 0.328\n"
          ]
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ],
            "image/png": "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\n"
          },
          "metadata": {
            "needs_background": "light"
          }
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Epoch 2\n",
            "624/624 [==============================] - 6s 9ms/step\n",
            "\n",
            "\n",
            "The best 0.346 at threshold 0.86 with precision 0.304 and recall 0.401\n"
          ]
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ],
            "image/png": "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\n"
          },
          "metadata": {
            "needs_background": "light"
          }
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Epoch 3\n",
            "624/624 [==============================] - 6s 9ms/step\n",
            "\n",
            "\n",
            "The best 0.326 at threshold 0.86 with precision 0.288 and recall 0.374\n"
          ]
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ],
            "image/png": "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\n"
          },
          "metadata": {
            "needs_background": "light"
          }
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Epoch 4\n",
            "624/624 [==============================] - 5s 9ms/step\n",
            "\n",
            "\n",
            "The best 0.344 at threshold 0.8200000000000001 with precision 0.321 and recall 0.371\n"
          ]
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ],
            "image/png": "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\n"
          },
          "metadata": {
            "needs_background": "light"
          }
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Epoch 5\n",
            "624/624 [==============================] - 5s 9ms/step\n",
            "\n",
            "\n",
            "The best 0.339 at threshold 0.8300000000000001 with precision 0.322 and recall 0.357\n"
          ]
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ],
            "image/png": "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\n"
          },
          "metadata": {
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "best_pd = pd.DataFrame(best_scores)\n",
        "print(best_pd)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "bDy9qDUWmEzX",
        "outputId": "da52a35c-3e7f-4ff1-8ee9-fd09a00d4c71"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "   thresh_best     f1_05   f1_best  precision_best  recall_best\n",
            "0         0.80  0.001468  0.329273        0.330861     0.327700\n",
            "1         0.86  0.194821  0.345922        0.304299     0.400735\n",
            "2         0.86  0.060863  0.325656        0.288385     0.373990\n",
            "3         0.82  0.259155  0.344123        0.320839     0.371051\n",
            "4         0.83  0.070621  0.338912        0.322495     0.357090\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "print(f\"Average best F1 is {np.mean(best_pd['f1_best']):.3f}\")"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "6vLrjjTkokyK",
        "outputId": "4ca15274-0faf-49a4-b3b0-ce8af98055ce"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Average best F1 is 0.337\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "It's slightly better than the best F1 we got from LogisticRegression and Naive Bayes, but the difference is as low as 0.01. Not a great deal.\n",
        "\n",
        "The interesting thing to observe is how different are the best_F1 values and F1 for 0.5 threshold on each fold. Different train set and different keras model weights initialization produced such a variety of results. Very unintuitive and instructive."
      ],
      "metadata": {
        "id": "gXc92nJNo14M"
      }
    },
    {
      "cell_type": "markdown",
      "source": [
        "This was the best result I managed to produce out of this dataset.\n",
        "\n",
        "However, it seemed tempting to see whether a state-of-the-art Natural Language Understanding model like BERT could give a better result? Is there any meaningful difference between expert and non-expert tweets that could be discovered by a contextual embedding? Let's try!\n",
        "\n",
        "**BERT model**\n",
        "\n",
        "We will use tweets processed with our data-cleaning procedure and TweetTokenizer as an input to BERT Tokenizer rather than raw text."
      ],
      "metadata": {
        "id": "ghpFJJhywPWR"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "pip install transformers"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "LKiCnBQHy-m-",
        "outputId": "50271f25-4ce8-4fd7-92ea-b24cb3cbdc1b"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Looking in indexes: https://pypi.org/simple, https://us-python.pkg.dev/colab-wheels/public/simple/\n",
            "Collecting transformers\n",
            "  Downloading transformers-4.23.1-py3-none-any.whl (5.3 MB)\n",
            "\u001b[K     |████████████████████████████████| 5.3 MB 25.7 MB/s \n",
            "\u001b[?25hRequirement already satisfied: tqdm>=4.27 in /usr/local/lib/python3.7/dist-packages (from transformers) (4.64.1)\n",
            "Requirement already satisfied: importlib-metadata in /usr/local/lib/python3.7/dist-packages (from transformers) (4.13.0)\n",
            "Requirement already satisfied: packaging>=20.0 in /usr/local/lib/python3.7/dist-packages (from transformers) (21.3)\n",
            "Requirement already satisfied: pyyaml>=5.1 in /usr/local/lib/python3.7/dist-packages (from transformers) (6.0)\n",
            "Collecting tokenizers!=0.11.3,<0.14,>=0.11.1\n",
            "  Downloading tokenizers-0.13.1-cp37-cp37m-manylinux_2_17_x86_64.manylinux2014_x86_64.whl (7.6 MB)\n",
            "\u001b[K     |████████████████████████████████| 7.6 MB 41.7 MB/s \n",
            "\u001b[?25hRequirement already satisfied: numpy>=1.17 in /usr/local/lib/python3.7/dist-packages (from transformers) (1.21.6)\n",
            "Requirement already satisfied: regex!=2019.12.17 in /usr/local/lib/python3.7/dist-packages (from transformers) (2022.6.2)\n",
            "Collecting huggingface-hub<1.0,>=0.10.0\n",
            "  Downloading huggingface_hub-0.10.1-py3-none-any.whl (163 kB)\n",
            "\u001b[K     |████████████████████████████████| 163 kB 42.7 MB/s \n",
            "\u001b[?25hRequirement already satisfied: filelock in /usr/local/lib/python3.7/dist-packages (from transformers) (3.8.0)\n",
            "Requirement already satisfied: requests in /usr/local/lib/python3.7/dist-packages (from transformers) (2.23.0)\n",
            "Requirement already satisfied: typing-extensions>=3.7.4.3 in /usr/local/lib/python3.7/dist-packages (from huggingface-hub<1.0,>=0.10.0->transformers) (4.1.1)\n",
            "Requirement already satisfied: pyparsing!=3.0.5,>=2.0.2 in /usr/local/lib/python3.7/dist-packages (from packaging>=20.0->transformers) (3.0.9)\n",
            "Requirement already satisfied: zipp>=0.5 in /usr/local/lib/python3.7/dist-packages (from importlib-metadata->transformers) (3.9.0)\n",
            "Requirement already satisfied: certifi>=2017.4.17 in /usr/local/lib/python3.7/dist-packages (from requests->transformers) (2022.9.24)\n",
            "Requirement already satisfied: idna<3,>=2.5 in /usr/local/lib/python3.7/dist-packages (from requests->transformers) (2.10)\n",
            "Requirement already satisfied: chardet<4,>=3.0.2 in /usr/local/lib/python3.7/dist-packages (from requests->transformers) (3.0.4)\n",
            "Requirement already satisfied: urllib3!=1.25.0,!=1.25.1,<1.26,>=1.21.1 in /usr/local/lib/python3.7/dist-packages (from requests->transformers) (1.24.3)\n",
            "Installing collected packages: tokenizers, huggingface-hub, transformers\n",
            "Successfully installed huggingface-hub-0.10.1 tokenizers-0.13.1 transformers-4.23.1\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "import pandas as pd\n",
        "import numpy as np\n",
        "import tensorflow as tf\n",
        "from transformers import BertTokenizer, TFBertForSequenceClassification\n",
        "tokenizer = BertTokenizer.from_pretrained('bert-base-uncased')\n",
        "from tensorflow.keras.optimizers.schedules import PolynomialDecay\n",
        "from keras.optimizers import Adam\n",
        "tf.config.run_functions_eagerly(True)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 113,
          "referenced_widgets": [
            "49d3cbe48b8a4d389c328ffb28268d34",
            "93c7a3c727d94661a55b16da5357ef01",
            "83bb56d48a1b43208db866d4766cd5d7",
            "09256a2b562d4caa93df6245f6969abf",
            "fe48db3d6f4140c4a5e2d8a2895cce43",
            "c69882ee465d477cbca0f1acca88d68d",
            "bf07759821524a808553b7834d68a989",
            "da0aecff3b3c4784b9a089bd01d08802",
            "a7e7c9860f484990af5673058da0148a",
            "76c2869f64924ae6b19f970e601a0d73",
            "814cfe825aa94b67aca149171c2a2b1d",
            "5a10256ef3ee44839e37f84a8400b7e5",
            "cbf63039110d4a7181900cea3fe02e81",
            "ccc9f853d1d74a9e9048647491ea78a3",
            "1c093f2473154b9aa219fa7617f96e52",
            "41ff7cbe96a341dcafba8d60902be9b1",
            "cd0f776c715646c3ab1aba589834a0c5",
            "40b260b7c75240efa8c73ab72cf51742",
            "ef579abff8d44263811bc4758412abfe",
            "8d7966ee53544cfb8daed844880bf74e",
            "161c3aaa29324c558a485531cc19b5ba",
            "c78ab646d95f49d88ceff2a484e4aa1e",
            "94c8d74ea72541218695bd3e8e1cb533",
            "b9257345b568489e8d4a4e4a6327d85e",
            "ff2a7cb1b7b14bc4ab2531947a4d9bd1",
            "73ec91fb2cd24e86b481319881fb18ca",
            "05745987b5e74b76b986a42fff94d988",
            "295e095abd2342658f3d7831828c0050",
            "b2b5e03887c244d9abcca2651df49615",
            "dd7ba37072b3442a9262e6028820f4c9",
            "83c6bc25306b4b9dbd5fa789c2eaeeda",
            "0f9c308c59824e11844db61b37773df7",
            "963b6d1d1e794ecfb090ea34a572fdf5"
          ]
        },
        "id": "6nk0g3syyoxT",
        "outputId": "c9b7802b-73cd-4471-bf6c-e17daa4b6c4b"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "Downloading:   0%|          | 0.00/232k [00:00<?, ?B/s]"
            ],
            "application/vnd.jupyter.widget-view+json": {
              "version_major": 2,
              "version_minor": 0,
              "model_id": "49d3cbe48b8a4d389c328ffb28268d34"
            }
          },
          "metadata": {}
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "Downloading:   0%|          | 0.00/28.0 [00:00<?, ?B/s]"
            ],
            "application/vnd.jupyter.widget-view+json": {
              "version_major": 2,
              "version_minor": 0,
              "model_id": "5a10256ef3ee44839e37f84a8400b7e5"
            }
          },
          "metadata": {}
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "Downloading:   0%|          | 0.00/570 [00:00<?, ?B/s]"
            ],
            "application/vnd.jupyter.widget-view+json": {
              "version_major": 2,
              "version_minor": 0,
              "model_id": "94c8d74ea72541218695bd3e8e1cb533"
            }
          },
          "metadata": {}
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "df = pd.read_csv('/content/drive/MyDrive/Python_data/processed_tweets.csv')\n",
        "df.dropna(inplace = True)\n",
        "df.reset_index(inplace = True)\n",
        "print(df.shape)\n",
        "y = df['expert']\n",
        "X = df['processed']"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "ml5bkDvDyxs7",
        "outputId": "0895e2a1-1908-49d0-b36f-8adc7c5a13e9"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "(99731, 11)\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "We use BERT for tensorflow in this project, so we have to put our data into a tf.data.Dataset object."
      ],
      "metadata": {
        "id": "o5ZW0NsqzwjV"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "X = X.tolist()\n",
        "X = tokenizer(X, padding = True, truncation = True, return_tensors = 'tf', max_length = 128)\n",
        "\n",
        "tf_dataset = tf.data.Dataset.from_tensor_slices((X.data, y))\n",
        "tf_dataset = tf_dataset.shuffle(100).batch(32)"
      ],
      "metadata": {
        "id": "wokYoo0PzJTp"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "test_size = int(len(tf_dataset) * 0.2)\n",
        "df_valid = tf_dataset.take(test_size)\n",
        "df_train = tf_dataset.skip(test_size)\n",
        "\n",
        "import tensorflow_datasets as tfds\n",
        "\n",
        "ytest = tfds.as_numpy(df_valid)\n",
        "ylabels = []\n",
        "for l in ytest:\n",
        "    ylabels.append(l[1])\n",
        "\n",
        "ylabels = np.concatenate(ylabels)"
      ],
      "metadata": {
        "id": "gR3TWjIwzR9l"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "source": [
        "We'll use the most popular bert-base-uncased checkpoint from a Huggingface library.\n",
        "\n",
        "We will also apply a smart learning rate scheduler, as a standard Adam learning rate is way too high for BERT model.\n",
        "\n",
        "To avoid my Colab notebook breaking, I train my BERT model on 1 epoch only. I tried it with more epochs, but the gain "
      ],
      "metadata": {
        "id": "6OAfhNwU5BLS"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "checkpoint = 'bert-base-uncased'\n",
        "model = TFBertForSequenceClassification.from_pretrained(checkpoint, num_labels = 2)\n",
        "\n",
        "num_epochs = 1\n",
        "num_train_steps = len(df_train) * num_epochs\n",
        "lr_scheduler = PolynomialDecay(initial_learning_rate = 5e-5, end_learning_rate = 0.0, decay_steps = num_train_steps)\n",
        "\n",
        "opt = Adam(learning_rate = lr_scheduler)\n",
        "model.compile(optimizer=opt,\n",
        "              loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True),\n",
        "              metrics=[tf.keras.metrics.SparseCategoricalAccuracy('accuracy')])\n",
        "\n",
        "print(model.summary())\n",
        "print(f\"Num train steps {num_train_steps}\")"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "LMdihV8FzhYP",
        "outputId": "4a6f4e54-7e4a-4da0-eadf-e45c137ea6a4"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "All model checkpoint layers were used when initializing TFBertForSequenceClassification.\n",
            "\n",
            "Some layers of TFBertForSequenceClassification were not initialized from the model checkpoint at bert-base-uncased and are newly initialized: ['classifier']\n",
            "You should probably TRAIN this model on a down-stream task to be able to use it for predictions and inference.\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Model: \"tf_bert_for_sequence_classification_2\"\n",
            "_________________________________________________________________\n",
            " Layer (type)                Output Shape              Param #   \n",
            "=================================================================\n",
            " bert (TFBertMainLayer)      multiple                  109482240 \n",
            "                                                                 \n",
            " dropout_113 (Dropout)       multiple                  0         \n",
            "                                                                 \n",
            " classifier (Dense)          multiple                  1538      \n",
            "                                                                 \n",
            "=================================================================\n",
            "Total params: 109,483,778\n",
            "Trainable params: 109,483,778\n",
            "Non-trainable params: 0\n",
            "_________________________________________________________________\n",
            "None\n",
            "Num train steps 2494\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "model.fit(df_train, validation_data = df_valid, verbose = 2, epochs = num_epochs)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "wdj1ma5uzuU6",
        "outputId": "721e6f83-13f2-4519-8025-c473deb33848"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "2494/2494 - 1807s - loss: 0.2267 - accuracy: 0.9325 - val_loss: 0.1934 - val_accuracy: 0.9422 - 1807s/epoch - 724ms/step\n"
          ]
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<keras.callbacks.History at 0x7f43f573a250>"
            ]
          },
          "metadata": {},
          "execution_count": 10
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "from sklearn.metrics import f1_score, log_loss, accuracy_score, recall_score, precision_score"
      ],
      "metadata": {
        "id": "gsmDNBaz0GTH"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "ypreds = model.predict(df_valid)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "9tQhpbFB2ZY1",
        "outputId": "668379b7-e8d4-4787-aeda-a3b8d84a1e29"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "623/623 [==============================] - 128s 206ms/step\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "ypreds = ypreds.logits\n",
        "print(ypreds[:5])"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "2W8iMjJd2gWZ",
        "outputId": "2062ead8-9206-41c4-a7e2-ed5c0ab515d2"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "[[ 0.46934596 -1.0455719 ]\n",
            " [ 0.78268856 -1.2551302 ]\n",
            " [ 1.6793638  -1.7786064 ]\n",
            " [ 0.46934596 -1.0455719 ]\n",
            " [ 1.2808208  -1.5485667 ]]\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "ypreds_soft = tf.math.softmax(ypreds).numpy()\n",
        "print(ypreds_soft[:5])"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "yhoNFVn_2hLk",
        "outputId": "25535013-a0fd-4c4d-c7d8-855b070a9953"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "[[0.81978893 0.18021111]\n",
            " [0.88471097 0.11528903]\n",
            " [0.9694679  0.03053206]\n",
            " [0.81978893 0.18021111]\n",
            " [0.9442434  0.05575662]]\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "ypreds_classes = tf.math.argmax(ypreds_soft, axis = -1)"
      ],
      "metadata": {
        "id": "q0KVqk7p3DqL"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "import matplotlib.pyplot as plt"
      ],
      "metadata": {
        "id": "ruHa5F5iH2Jx"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "threhsold_f1(ypreds_soft[:, 0], ylabels)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 333
        },
        "id": "l8K_E_a13QHN",
        "outputId": "47988267-d096-4778-a578-6b965cd8a403"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "\n",
            "The best 0.117 at threshold 0.98 with precision 0.062 and recall 1.000\n"
          ]
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ],
            "image/png": "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\n"
          },
          "metadata": {
            "needs_background": "light"
          }
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "(0.98, 0.11729152894513174, 0.06229935794542536, 1.0)"
            ]
          },
          "metadata": {},
          "execution_count": 24
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "It looks like the difference between expert and non-expert tweets does not lie in their \"meaning\". A contextual embedding from BERT produced by far the worst result. Which doesn't change the fact that generally BERT rocks!"
      ],
      "metadata": {
        "id": "qdMigQPDIJVG"
      }
    },
    {
      "cell_type": "markdown",
      "source": [
        "This is the end of the notebook. I hope you had fun :-)"
      ],
      "metadata": {
        "id": "aGnEh84bIdC3"
      }
    }
  ]
}