ResNeXt with CIFAR10.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "f1b203b8",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "['/device:CPU:0', '/device:GPU:0']\n"
     ]
    }
   ],
   "source": [
    "%matplotlib inline\n",
    "%load_ext autoreload\n",
    "%autoreload 2\n",
    "import tensorflow as tf\n",
    "from tensorflow.keras.models import Model\n",
    "from tensorflow.compat.v1 import ConfigProto\n",
    "from tensorflow.compat.v1 import InteractiveSession\n",
    "\n",
    "config = ConfigProto()\n",
    "config.gpu_options.allow_growth = True\n",
    "session = InteractiveSession(config=config)\n",
    "from tensorflow.python.client import device_lib\n",
    "def get_available_devices():\n",
    "    local_device_protos = device_lib.list_local_devices()\n",
    "    return [x.name for x in local_device_protos]\n",
    "print(get_available_devices()) \n",
    "\n",
    "import os\n",
    "import pandas as pd\n",
    "import numpy as np\n",
    "from PIL import Image\n",
    "import matplotlib.pyplot as plt\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "e1ac925a",
   "metadata": {},
   "source": [
    "# 0. Loading the data"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "e0c35132",
   "metadata": {},
   "source": [
    "### TinyImageNet does not have labels for the testing data. We will use the 10k validation images for the testing data."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "513cdce2",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(50000, 32, 32, 3)\n",
      "(50000, 10)\n",
      "(10000, 32, 32, 3)\n",
      "(10000, 10)\n"
     ]
    }
   ],
   "source": [
    "from tensorflow.keras.utils import to_categorical  \n",
    "\n",
    "#load cifar-10 dataset\n",
    "(X_tot, y_tot), (X_test, y_test) = tf.keras.datasets.cifar10.load_data()\n",
    "#one hot encoding\n",
    "y_test=to_categorical(y_test, num_classes=10)\n",
    "y_tot=to_categorical(y_tot, num_classes=10)\n",
    "\n",
    "print(X_tot.shape)\n",
    "print(y_tot.shape)\n",
    "print(X_test.shape)\n",
    "print(y_test.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "1df1e044",
   "metadata": {},
   "outputs": [],
   "source": [
    "#shuffle data and split data to generate validation data\n",
    "shuffler = np.random.RandomState(seed=42).permutation(len(X_tot))\n",
    "X_tot = X_tot[shuffler]\n",
    "y_tot = y_tot[shuffler]\n",
    " \n",
    "\n",
    "#validation data\n",
    "X_val=X_tot[0:2000]\n",
    "y_val=y_tot[0:2000]\n",
    "\n",
    "X_train=X_tot[2000:]\n",
    "y_train=y_tot[2000:]"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "ebc758bc",
   "metadata": {},
   "source": [
    "# 1. Using ResNeXt29_8x64d with CIFAR10"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "c7173b94",
   "metadata": {},
   "source": [
    "### We use our custom resnet model generator to get any resnet \n",
    "\n",
    "CIFAR 10 has 10 different labels"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "aab86022",
   "metadata": {},
   "source": [
    "We use a custom resnet model, because the images are 32x32 we do not need to go as deep to get good results.The CNN layers go to 2x2 resolution instead to 1x1 in a traditional resnet model."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "375a049b",
   "metadata": {},
   "outputs": [],
   "source": [
    "from utils.custom_ResNeXt import custom_ResNeXt, custom_ResNeXt29\n",
    "\n",
    "ResNeXt29=custom_ResNeXt29(input_shape=(32,32,3),n_classes=10,cardinality=8,model_name='custom_ResNext29_8x64d')\n",
    "#tf.keras.utils.plot_model(ResNeXt29, to_file='test.png',show_shapes=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "4a4c82cd-5906-4810-8735-ad99046ff483",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/200\n",
      "375/375 [==============================] - 82s 180ms/step - loss: 2.1490 - acc: 0.3136 - val_loss: 1.4490 - val_acc: 0.4705\n",
      "\n",
      "Epoch 00001: val_acc improved from -inf to 0.47050, saving model to ./models/cifar10/ResNext29_best_cifar10.hdf5\n",
      "Epoch 2/200\n",
      "375/375 [==============================] - 65s 173ms/step - loss: 1.3568 - acc: 0.5243 - val_loss: 1.2632 - val_acc: 0.5470\n",
      "\n",
      "Epoch 00002: val_acc improved from 0.47050 to 0.54700, saving model to ./models/cifar10/ResNext29_best_cifar10.hdf5\n",
      "Epoch 3/200\n",
      "375/375 [==============================] - 65s 172ms/step - loss: 1.1392 - acc: 0.6000 - val_loss: 1.0263 - val_acc: 0.6395\n",
      "\n",
      "Epoch 00003: val_acc improved from 0.54700 to 0.63950, saving model to ./models/cifar10/ResNext29_best_cifar10.hdf5\n",
      "Epoch 4/200\n",
      "375/375 [==============================] - 65s 172ms/step - loss: 1.0083 - acc: 0.6474 - val_loss: 1.2370 - val_acc: 0.5700\n",
      "\n",
      "Epoch 00004: val_acc did not improve from 0.63950\n",
      "Epoch 5/200\n",
      "375/375 [==============================] - 64s 171ms/step - loss: 0.9124 - acc: 0.6789 - val_loss: 0.8813 - val_acc: 0.6865\n",
      "\n",
      "Epoch 00005: val_acc improved from 0.63950 to 0.68650, saving model to ./models/cifar10/ResNext29_best_cifar10.hdf5\n",
      "Epoch 6/200\n",
      "375/375 [==============================] - 64s 171ms/step - loss: 0.8218 - acc: 0.7113 - val_loss: 0.8269 - val_acc: 0.7045\n",
      "\n",
      "Epoch 00006: val_acc improved from 0.68650 to 0.70450, saving model to ./models/cifar10/ResNext29_best_cifar10.hdf5\n",
      "Epoch 7/200\n",
      "375/375 [==============================] - 65s 172ms/step - loss: 0.7487 - acc: 0.7413 - val_loss: 1.0109 - val_acc: 0.6775\n",
      "\n",
      "Epoch 00007: val_acc did not improve from 0.70450\n",
      "Epoch 8/200\n",
      "375/375 [==============================] - 64s 172ms/step - loss: 0.6904 - acc: 0.7579 - val_loss: 0.8984 - val_acc: 0.7030\n",
      "\n",
      "Epoch 00008: val_acc did not improve from 0.70450\n",
      "Epoch 9/200\n",
      "375/375 [==============================] - 64s 171ms/step - loss: 0.6489 - acc: 0.7735 - val_loss: 0.7726 - val_acc: 0.7300\n",
      "\n",
      "Epoch 00009: val_acc improved from 0.70450 to 0.73000, saving model to ./models/cifar10/ResNext29_best_cifar10.hdf5\n",
      "Epoch 10/200\n",
      "375/375 [==============================] - 64s 171ms/step - loss: 0.6180 - acc: 0.7844 - val_loss: 0.8335 - val_acc: 0.7115\n",
      "\n",
      "Epoch 00010: val_acc did not improve from 0.73000\n",
      "Epoch 11/200\n",
      "375/375 [==============================] - 64s 172ms/step - loss: 0.5740 - acc: 0.8019 - val_loss: 0.8303 - val_acc: 0.7265\n",
      "\n",
      "Epoch 00011: val_acc did not improve from 0.73000\n",
      "Epoch 12/200\n",
      "375/375 [==============================] - 64s 172ms/step - loss: 0.5382 - acc: 0.8119 - val_loss: 0.6937 - val_acc: 0.7620\n",
      "\n",
      "Epoch 00012: val_acc improved from 0.73000 to 0.76200, saving model to ./models/cifar10/ResNext29_best_cifar10.hdf5\n",
      "Epoch 13/200\n",
      "375/375 [==============================] - 64s 171ms/step - loss: 0.5125 - acc: 0.8201 - val_loss: 0.7138 - val_acc: 0.7525\n",
      "\n",
      "Epoch 00013: val_acc did not improve from 0.76200\n",
      "Epoch 14/200\n",
      "375/375 [==============================] - 64s 171ms/step - loss: 0.4790 - acc: 0.8310 - val_loss: 0.7456 - val_acc: 0.7570\n",
      "\n",
      "Epoch 00014: val_acc did not improve from 0.76200\n",
      "Epoch 15/200\n",
      "375/375 [==============================] - 64s 171ms/step - loss: 0.4613 - acc: 0.8386 - val_loss: 0.6441 - val_acc: 0.7790\n",
      "\n",
      "Epoch 00015: val_acc improved from 0.76200 to 0.77900, saving model to ./models/cifar10/ResNext29_best_cifar10.hdf5\n",
      "Epoch 16/200\n",
      "375/375 [==============================] - 64s 171ms/step - loss: 0.4262 - acc: 0.8507 - val_loss: 0.6558 - val_acc: 0.7765\n",
      "\n",
      "Epoch 00016: val_acc did not improve from 0.77900\n",
      "Epoch 17/200\n",
      "375/375 [==============================] - 64s 171ms/step - loss: 0.3934 - acc: 0.8638 - val_loss: 0.6013 - val_acc: 0.7965\n",
      "\n",
      "Epoch 00017: val_acc improved from 0.77900 to 0.79650, saving model to ./models/cifar10/ResNext29_best_cifar10.hdf5\n",
      "Epoch 18/200\n",
      "375/375 [==============================] - 64s 171ms/step - loss: 0.3830 - acc: 0.8688 - val_loss: 0.6139 - val_acc: 0.7985\n",
      "\n",
      "Epoch 00018: val_acc improved from 0.79650 to 0.79850, saving model to ./models/cifar10/ResNext29_best_cifar10.hdf5\n",
      "Epoch 19/200\n",
      "375/375 [==============================] - 64s 171ms/step - loss: 0.3691 - acc: 0.8700 - val_loss: 0.7163 - val_acc: 0.7660\n",
      "\n",
      "Epoch 00019: val_acc did not improve from 0.79850\n",
      "Epoch 20/200\n",
      "375/375 [==============================] - 64s 171ms/step - loss: 0.3439 - acc: 0.8817 - val_loss: 0.7412 - val_acc: 0.7690\n",
      "\n",
      "Epoch 00020: val_acc did not improve from 0.79850\n",
      "Epoch 21/200\n",
      "375/375 [==============================] - 64s 171ms/step - loss: 0.3342 - acc: 0.8807 - val_loss: 0.6206 - val_acc: 0.7880\n",
      "\n",
      "Epoch 00021: val_acc did not improve from 0.79850\n",
      "Epoch 22/200\n",
      "375/375 [==============================] - 64s 171ms/step - loss: 0.3138 - acc: 0.8900 - val_loss: 0.5581 - val_acc: 0.8165\n",
      "\n",
      "Epoch 00022: val_acc improved from 0.79850 to 0.81650, saving model to ./models/cifar10/ResNext29_best_cifar10.hdf5\n",
      "Epoch 23/200\n",
      "375/375 [==============================] - 64s 171ms/step - loss: 0.2897 - acc: 0.8974 - val_loss: 0.7170 - val_acc: 0.7795\n",
      "\n",
      "Epoch 00023: val_acc did not improve from 0.81650\n",
      "Epoch 24/200\n",
      "375/375 [==============================] - 64s 171ms/step - loss: 0.2783 - acc: 0.9026 - val_loss: 0.5689 - val_acc: 0.8285\n",
      "\n",
      "Epoch 00024: val_acc improved from 0.81650 to 0.82850, saving model to ./models/cifar10/ResNext29_best_cifar10.hdf5\n",
      "Epoch 25/200\n",
      "375/375 [==============================] - 64s 171ms/step - loss: 0.2658 - acc: 0.9084 - val_loss: 0.7599 - val_acc: 0.7860\n",
      "\n",
      "Epoch 00025: val_acc did not improve from 0.82850\n",
      "Epoch 26/200\n",
      "375/375 [==============================] - 64s 171ms/step - loss: 0.2554 - acc: 0.9117 - val_loss: 0.5608 - val_acc: 0.8245\n",
      "\n",
      "Epoch 00026: val_acc did not improve from 0.82850\n",
      "Epoch 27/200\n",
      "375/375 [==============================] - 64s 171ms/step - loss: 0.2394 - acc: 0.9157 - val_loss: 0.9137 - val_acc: 0.7540\n",
      "\n",
      "Epoch 00027: val_acc did not improve from 0.82850\n",
      "Epoch 28/200\n",
      "375/375 [==============================] - 64s 171ms/step - loss: 0.2302 - acc: 0.9212 - val_loss: 0.6631 - val_acc: 0.7950\n",
      "\n",
      "Epoch 00028: val_acc did not improve from 0.82850\n",
      "Epoch 29/200\n",
      "375/375 [==============================] - 64s 171ms/step - loss: 0.2183 - acc: 0.9218 - val_loss: 0.6458 - val_acc: 0.8090\n",
      "\n",
      "Epoch 00029: ReduceLROnPlateau reducing learning rate to 0.001.\n",
      "\n",
      "Epoch 00029: val_acc did not improve from 0.82850\n",
      "Epoch 30/200\n",
      "375/375 [==============================] - 64s 171ms/step - loss: 0.1923 - acc: 0.9320 - val_loss: 0.4740 - val_acc: 0.8535\n",
      "\n",
      "Epoch 00030: val_acc improved from 0.82850 to 0.85350, saving model to ./models/cifar10/ResNext29_best_cifar10.hdf5\n",
      "Epoch 31/200\n",
      "375/375 [==============================] - 64s 171ms/step - loss: 0.1288 - acc: 0.9574 - val_loss: 0.4873 - val_acc: 0.8500\n",
      "\n",
      "Epoch 00031: val_acc did not improve from 0.85350\n",
      "Epoch 32/200\n",
      "375/375 [==============================] - 64s 171ms/step - loss: 0.1199 - acc: 0.9605 - val_loss: 0.4902 - val_acc: 0.8545\n",
      "\n",
      "Epoch 00032: val_acc improved from 0.85350 to 0.85450, saving model to ./models/cifar10/ResNext29_best_cifar10.hdf5\n",
      "Epoch 33/200\n",
      "375/375 [==============================] - 64s 171ms/step - loss: 0.1100 - acc: 0.9625 - val_loss: 0.4961 - val_acc: 0.8565\n",
      "\n",
      "Epoch 00033: val_acc improved from 0.85450 to 0.85650, saving model to ./models/cifar10/ResNext29_best_cifar10.hdf5\n",
      "Epoch 34/200\n",
      "375/375 [==============================] - 64s 171ms/step - loss: 0.1066 - acc: 0.9643 - val_loss: 0.5047 - val_acc: 0.8515\n",
      "\n",
      "Epoch 00034: val_acc did not improve from 0.85650\n",
      "Epoch 35/200\n",
      "375/375 [==============================] - 64s 171ms/step - loss: 0.1009 - acc: 0.9666 - val_loss: 0.5181 - val_acc: 0.8525\n",
      "\n",
      "Epoch 00035: val_acc did not improve from 0.85650\n",
      "Epoch 36/200\n",
      "375/375 [==============================] - 64s 171ms/step - loss: 0.0897 - acc: 0.9688 - val_loss: 0.5115 - val_acc: 0.8515\n",
      "\n",
      "Epoch 00036: val_acc did not improve from 0.85650\n",
      "Epoch 37/200\n",
      "375/375 [==============================] - 64s 171ms/step - loss: 0.0886 - acc: 0.9703 - val_loss: 0.5246 - val_acc: 0.8520\n",
      "\n",
      "Epoch 00037: val_acc did not improve from 0.85650\n",
      "Epoch 38/200\n",
      "375/375 [==============================] - 64s 171ms/step - loss: 0.0840 - acc: 0.9730 - val_loss: 0.5116 - val_acc: 0.8545\n",
      "\n",
      "Epoch 00038: ReduceLROnPlateau reducing learning rate to 0.001.\n",
      "\n",
      "Epoch 00038: val_acc did not improve from 0.85650\n",
      "Epoch 39/200\n",
      "375/375 [==============================] - 64s 171ms/step - loss: 0.0798 - acc: 0.9733 - val_loss: 0.5224 - val_acc: 0.8575\n",
      "\n",
      "Epoch 00039: val_acc improved from 0.85650 to 0.85750, saving model to ./models/cifar10/ResNext29_best_cifar10.hdf5\n",
      "Epoch 40/200\n",
      "375/375 [==============================] - 64s 172ms/step - loss: 0.0767 - acc: 0.9747 - val_loss: 0.5423 - val_acc: 0.8480\n",
      "\n",
      "Epoch 00040: val_acc did not improve from 0.85750\n",
      "Epoch 41/200\n",
      "375/375 [==============================] - 64s 171ms/step - loss: 0.0751 - acc: 0.9753 - val_loss: 0.5408 - val_acc: 0.8500\n",
      "\n",
      "Epoch 00041: val_acc did not improve from 0.85750\n",
      "Epoch 42/200\n",
      "375/375 [==============================] - 64s 172ms/step - loss: 0.0716 - acc: 0.9750 - val_loss: 0.5549 - val_acc: 0.8540\n",
      "\n",
      "Epoch 00042: val_acc did not improve from 0.85750\n",
      "Epoch 43/200\n",
      "375/375 [==============================] - 64s 171ms/step - loss: 0.0681 - acc: 0.9776 - val_loss: 0.5613 - val_acc: 0.8505\n",
      "\n",
      "Epoch 00043: val_acc did not improve from 0.85750\n",
      "Epoch 44/200\n",
      "375/375 [==============================] - 64s 171ms/step - loss: 0.0705 - acc: 0.9763 - val_loss: 0.5669 - val_acc: 0.8500\n",
      "\n",
      "Epoch 00044: ReduceLROnPlateau reducing learning rate to 0.001.\n",
      "\n",
      "Epoch 00044: val_acc did not improve from 0.85750\n",
      "Epoch 45/200\n",
      "375/375 [==============================] - 64s 171ms/step - loss: 0.0703 - acc: 0.9769 - val_loss: 0.5716 - val_acc: 0.8525\n",
      "\n",
      "Epoch 00045: val_acc did not improve from 0.85750\n",
      "Epoch 46/200\n",
      "375/375 [==============================] - 64s 171ms/step - loss: 0.0695 - acc: 0.9766 - val_loss: 0.5508 - val_acc: 0.8565\n",
      "\n",
      "Epoch 00046: val_acc did not improve from 0.85750\n",
      "Epoch 47/200\n",
      "375/375 [==============================] - 64s 171ms/step - loss: 0.0617 - acc: 0.9799 - val_loss: 0.5649 - val_acc: 0.8565\n",
      "\n",
      "Epoch 00047: val_acc did not improve from 0.85750\n",
      "Epoch 48/200\n",
      "375/375 [==============================] - 64s 171ms/step - loss: 0.0603 - acc: 0.9791 - val_loss: 0.5585 - val_acc: 0.8515\n",
      "\n",
      "Epoch 00048: val_acc did not improve from 0.85750\n",
      "Epoch 49/200\n",
      "375/375 [==============================] - 64s 171ms/step - loss: 0.0579 - acc: 0.9816 - val_loss: 0.5743 - val_acc: 0.8555\n",
      "\n",
      "Epoch 00049: ReduceLROnPlateau reducing learning rate to 0.001.\n",
      "\n",
      "Epoch 00049: val_acc did not improve from 0.85750\n",
      "Epoch 50/200\n",
      "375/375 [==============================] - 64s 171ms/step - loss: 0.0563 - acc: 0.9817 - val_loss: 0.5679 - val_acc: 0.8620\n",
      "\n",
      "Epoch 00050: val_acc improved from 0.85750 to 0.86200, saving model to ./models/cifar10/ResNext29_best_cifar10.hdf5\n",
      "Epoch 51/200\n",
      "375/375 [==============================] - 64s 171ms/step - loss: 0.0557 - acc: 0.9814 - val_loss: 0.5834 - val_acc: 0.8570\n",
      "\n",
      "Epoch 00051: val_acc did not improve from 0.86200\n",
      "Epoch 52/200\n",
      "375/375 [==============================] - 64s 171ms/step - loss: 0.0536 - acc: 0.9825 - val_loss: 0.5929 - val_acc: 0.8555\n",
      "\n",
      "Epoch 00052: val_acc did not improve from 0.86200\n",
      "Epoch 53/200\n",
      "375/375 [==============================] - 64s 170ms/step - loss: 0.0550 - acc: 0.9819 - val_loss: 0.5815 - val_acc: 0.8600\n",
      "\n",
      "Epoch 00053: val_acc did not improve from 0.86200\n",
      "Epoch 54/200\n",
      "375/375 [==============================] - 64s 171ms/step - loss: 0.0529 - acc: 0.9830 - val_loss: 0.5829 - val_acc: 0.8545\n",
      "\n",
      "Epoch 00054: val_acc did not improve from 0.86200\n",
      "Epoch 55/200\n",
      "375/375 [==============================] - 64s 170ms/step - loss: 0.0499 - acc: 0.9830 - val_loss: 0.5849 - val_acc: 0.8565\n",
      "\n",
      "Epoch 00055: ReduceLROnPlateau reducing learning rate to 0.001.\n",
      "\n",
      "Epoch 00055: val_acc did not improve from 0.86200\n",
      "Epoch 56/200\n",
      "375/375 [==============================] - 64s 170ms/step - loss: 0.0516 - acc: 0.9828 - val_loss: 0.5969 - val_acc: 0.8600\n",
      "\n",
      "Epoch 00056: val_acc did not improve from 0.86200\n",
      "Epoch 57/200\n",
      "375/375 [==============================] - 64s 171ms/step - loss: 0.0468 - acc: 0.9846 - val_loss: 0.6024 - val_acc: 0.8525\n",
      "\n",
      "Epoch 00057: val_acc did not improve from 0.86200\n",
      "Epoch 58/200\n",
      "375/375 [==============================] - 64s 171ms/step - loss: 0.0476 - acc: 0.9847 - val_loss: 0.6100 - val_acc: 0.8540\n",
      "\n",
      "Epoch 00058: val_acc did not improve from 0.86200\n",
      "Epoch 59/200\n",
      "375/375 [==============================] - 64s 171ms/step - loss: 0.0454 - acc: 0.9847 - val_loss: 0.6089 - val_acc: 0.8550\n",
      "\n",
      "Epoch 00059: val_acc did not improve from 0.86200\n",
      "Epoch 60/200\n",
      "375/375 [==============================] - 64s 170ms/step - loss: 0.0493 - acc: 0.9830 - val_loss: 0.6141 - val_acc: 0.8535\n",
      "\n",
      "Epoch 00060: ReduceLROnPlateau reducing learning rate to 0.001.\n",
      "\n",
      "Epoch 00060: val_acc did not improve from 0.86200\n",
      "Epoch 61/200\n",
      "375/375 [==============================] - 64s 171ms/step - loss: 0.0410 - acc: 0.9865 - val_loss: 0.6261 - val_acc: 0.8570\n",
      "\n",
      "Epoch 00061: val_acc did not improve from 0.86200\n",
      "Epoch 62/200\n",
      "375/375 [==============================] - 64s 171ms/step - loss: 0.0413 - acc: 0.9861 - val_loss: 0.6281 - val_acc: 0.8555\n",
      "\n",
      "Epoch 00062: val_acc did not improve from 0.86200\n",
      "Epoch 63/200\n",
      "375/375 [==============================] - 64s 171ms/step - loss: 0.0396 - acc: 0.9864 - val_loss: 0.6291 - val_acc: 0.8510\n",
      "\n",
      "Epoch 00063: val_acc did not improve from 0.86200\n",
      "Epoch 64/200\n",
      "375/375 [==============================] - 64s 171ms/step - loss: 0.0393 - acc: 0.9870 - val_loss: 0.6443 - val_acc: 0.8530\n",
      "\n",
      "Epoch 00064: val_acc did not improve from 0.86200\n",
      "Epoch 65/200\n",
      "375/375 [==============================] - 64s 171ms/step - loss: 0.0362 - acc: 0.9883 - val_loss: 0.6400 - val_acc: 0.8560\n",
      "\n",
      "Epoch 00065: ReduceLROnPlateau reducing learning rate to 0.001.\n",
      "\n",
      "Epoch 00065: val_acc did not improve from 0.86200\n"
     ]
    }
   ],
   "source": [
    "from tensorflow.keras.models import Model\n",
    "from tensorflow.keras.preprocessing.image import ImageDataGenerator\n",
    "from tensorflow.keras.callbacks import LearningRateScheduler,ModelCheckpoint,ReduceLROnPlateau\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "path='./models/cifar10/ResNext29_best_cifar10.hdf5'\n",
    "train_datagen = ImageDataGenerator(\n",
    "        horizontal_flip=True,\n",
    "        width_shift_range=4, #+-8 pixel shift max\n",
    "        height_shift_range=4)#+-8 pixel shift max\n",
    "\n",
    "\n",
    "reduce_lr = ReduceLROnPlateau(monitor='val_acc', mode='max',factor=0.1,\n",
    "                              patience=5, min_lr=0.001, verbose=1)\n",
    "\n",
    "#Early stopping\n",
    "es = tf.keras.callbacks.EarlyStopping(monitor='val_acc', mode='max', patience=15)\n",
    "# checkpoints\n",
    "checkpoint = ModelCheckpoint(path, monitor='val_acc', verbose=1, save_best_only=True, mode='max')\n",
    "ResNeXt29.compile(optimizer=tf.keras.optimizers.SGD(learning_rate=0.01,momentum=0.9),loss='categorical_crossentropy', metrics = ['acc'])\n",
    "\n",
    "\n",
    "train_generator = train_datagen.flow(X_train, y_train, batch_size=128)\n",
    "step_size_train=train_generator.n//train_generator.batch_size\n",
    "\n",
    "history=ResNeXt29.fit(train_generator,\n",
    "                    steps_per_epoch=step_size_train,\n",
    "                    epochs=200,\n",
    "                    validation_data=(X_val,y_val),\n",
    "                    callbacks=[reduce_lr,es,checkpoint])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "5837322e-e254-4d70-b344-6d4932c11042",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 576x576 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "from utils.train_CIFAR_all import plot_model\n",
    "plot_model(history)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "f4e9ef62-a26d-49ce-af81-689385f7f698",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[0.3959791660308838, 0.5465208292007446, 0.6069583296775818, 0.6514999866485596, 0.6833750009536743, 0.7103124856948853, 0.7362916469573975, 0.7568125128746033, 0.7724166512489319, 0.7828541398048401, 0.7975416779518127, 0.8066250085830688, 0.8186874985694885, 0.8306041955947876, 0.839062511920929, 0.8467291593551636, 0.8571666479110718, 0.8636875152587891, 0.8687916398048401, 0.8784999847412109, 0.8792708516120911, 0.885854184627533, 0.8940625190734863, 0.8996875286102295, 0.9036666750907898, 0.9068958163261414, 0.9125208258628845, 0.917229175567627, 0.9182083606719971, 0.9426875114440918, 0.9580625295639038, 0.9603124856948853, 0.9632291793823242, 0.9647708535194397, 0.9664999842643738, 0.9684374928474426, 0.9704166650772095, 0.9715833067893982, 0.9732916951179504, 0.9736666679382324, 0.9739791750907898, 0.9754999876022339, 0.9779999852180481, 0.9762916564941406, 0.9779791831970215, 0.9777500033378601, 0.9799374938011169, 0.9794166684150696, 0.9807083606719971, 0.9812708497047424, 0.9816250205039978, 0.9827083349227905, 0.9826250076293945, 0.9824583530426025, 0.9837499856948853, 0.9830625057220459, 0.9836041927337646, 0.9849374890327454, 0.9841874837875366, 0.9848958253860474, 0.9863749742507935, 0.9857708215713501, 0.9864583611488342, 0.9866874814033508, 0.9867708086967468]\n",
      "[1.8104857206344604, 1.302119493484497, 1.1200172901153564, 0.9972066283226013, 0.8981835842132568, 0.8224405646324158, 0.7602211833000183, 0.6968370079994202, 0.654060959815979, 0.624396800994873, 0.5836455821990967, 0.5528632998466492, 0.5147693157196045, 0.481920450925827, 0.46225109696388245, 0.4368053376674652, 0.4117490351200104, 0.39263200759887695, 0.37531399726867676, 0.3495720624923706, 0.3412160575389862, 0.3270028531551361, 0.3029703199863434, 0.287662148475647, 0.27787938714027405, 0.26950782537460327, 0.24884521961212158, 0.23998695611953735, 0.2318170815706253, 0.16623003780841827, 0.12748688459396362, 0.11851515620946884, 0.10893479734659195, 0.10410606861114502, 0.0996474102139473, 0.09211984276771545, 0.08984160423278809, 0.086410753428936, 0.08059310168027878, 0.07862065732479095, 0.07758812606334686, 0.07233928889036179, 0.06829136610031128, 0.06958132237195969, 0.06883106380701065, 0.06758839637041092, 0.06279028952121735, 0.06036577746272087, 0.058853525668382645, 0.05673075467348099, 0.054731566458940506, 0.05454486235976219, 0.05236004665493965, 0.05444170534610748, 0.04959268122911453, 0.05003902316093445, 0.04962543398141861, 0.046556681394577026, 0.04683849588036537, 0.045036498457193375, 0.04221128672361374, 0.04192013666033745, 0.04068409651517868, 0.0410434752702713, 0.03966309130191803]\n",
      "[0.47049999237060547, 0.546999990940094, 0.6395000219345093, 0.5699999928474426, 0.6865000128746033, 0.7045000195503235, 0.6775000095367432, 0.703000009059906, 0.7300000190734863, 0.7114999890327454, 0.7264999747276306, 0.7620000243186951, 0.7524999976158142, 0.7570000290870667, 0.7789999842643738, 0.7764999866485596, 0.796500027179718, 0.7985000014305115, 0.765999972820282, 0.7689999938011169, 0.7879999876022339, 0.8165000081062317, 0.7795000076293945, 0.828499972820282, 0.7860000133514404, 0.8245000243186951, 0.7540000081062317, 0.7950000166893005, 0.8090000152587891, 0.8535000085830688, 0.8500000238418579, 0.8544999957084656, 0.8565000295639038, 0.8514999747276306, 0.8525000214576721, 0.8514999747276306, 0.8519999980926514, 0.8544999957084656, 0.8575000166893005, 0.8479999899864197, 0.8500000238418579, 0.8539999723434448, 0.8504999876022339, 0.8500000238418579, 0.8525000214576721, 0.8565000295639038, 0.8565000295639038, 0.8514999747276306, 0.8554999828338623, 0.8619999885559082, 0.8569999933242798, 0.8554999828338623, 0.8600000143051147, 0.8544999957084656, 0.8565000295639038, 0.8600000143051147, 0.8525000214576721, 0.8539999723434448, 0.8550000190734863, 0.8535000085830688, 0.8569999933242798, 0.8554999828338623, 0.8510000109672546, 0.8529999852180481, 0.8560000061988831]\n",
      "[1.4490455389022827, 1.2631957530975342, 1.0263431072235107, 1.2369827032089233, 0.8813456892967224, 0.8269264101982117, 1.0108637809753418, 0.8984369039535522, 0.7725762724876404, 0.8335421681404114, 0.8302772045135498, 0.6937455534934998, 0.7138118147850037, 0.7456356287002563, 0.6441232562065125, 0.6558362245559692, 0.6013484001159668, 0.6138559579849243, 0.7163292765617371, 0.7411890625953674, 0.6206064224243164, 0.5580945611000061, 0.7170336842536926, 0.5689497590065002, 0.7599480748176575, 0.5608351826667786, 0.9136954545974731, 0.6631442904472351, 0.64583820104599, 0.4739511013031006, 0.48734962940216064, 0.4901899993419647, 0.49606195092201233, 0.5046903491020203, 0.5181359648704529, 0.5114510655403137, 0.5246211886405945, 0.5116098523139954, 0.5223538279533386, 0.5423475503921509, 0.5407730340957642, 0.5548572540283203, 0.5612667202949524, 0.5669481158256531, 0.5715510249137878, 0.55079585313797, 0.5648751258850098, 0.5584904551506042, 0.5742747783660889, 0.5679301619529724, 0.5833954215049744, 0.5929497480392456, 0.5814514756202698, 0.5828905701637268, 0.5849300026893616, 0.5969416499137878, 0.6024125218391418, 0.6100291609764099, 0.6088725924491882, 0.6140758395195007, 0.6260855793952942, 0.6281145811080933, 0.6291131377220154, 0.644278347492218, 0.6399767398834229]\n"
     ]
    }
   ],
   "source": [
    "print(history.history['acc'])\n",
    "print(history.history['loss'])\n",
    "print(history.history['val_acc'])\n",
    "print(history.history['val_loss'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "id": "a9d84aa5-b704-40c9-89ff-ad9e3c87a7ef",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "313/313 - 13s - loss: 0.7287 - acc: 0.8287\n",
      "custom_ResNext29_8x64d\n",
      "model accuracy: 82.87%\n",
      "top 3 accuracy: 96.52000069618225\n",
      "top 5 accuracy: 99.08999800682068\n"
     ]
    }
   ],
   "source": [
    "from utils.evaluate_model import score\n",
    "# reload from saved weights:\n",
    "model=custom_ResNeXt29(input_shape=(32,32,3),n_classes=10,cardinality=8,model_name='custom_ResNext29_8x64d')\n",
    "model.load_weights(path)\n",
    "\n",
    "if not os.path.exists('./full_models/cifar10'):\n",
    "    os.makedirs('./full_models/cifar10')\n",
    "model.save('./full_models/cifar10/ResNext29_best.h5')\n",
    "\n",
    "\n",
    "score(model,X_test,y_test,top_n=[3,5])"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "9cae9b57-3cd8-47d2-af0f-93367a76d8a1",
   "metadata": {},
   "source": [
    "# 2. Using SE_ResNeXt29_8x64d with CIFAR10"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "id": "a7e104cd-91d5-4db6-8cc3-b6269f953480",
   "metadata": {},
   "outputs": [],
   "source": [
    "from utils.SE_ResNeXt import SE_ResNeXt29\n",
    "\n",
    "SEResNeXt29=SE_ResNeXt29(input_shape=(32,32,3),n_classes=10,ratio=16,name='custom_ResNeXt29_SE')\n",
    "#SEResNeXt.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "id": "297ffe40-6b01-4982-bcac-70c5424417a7",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/200\n",
      "375/375 [==============================] - 82s 189ms/step - loss: 1.8576 - acc: 0.3535 - val_loss: 1.4103 - val_acc: 0.4970\n",
      "\n",
      "Epoch 00001: val_acc improved from -inf to 0.49700, saving model to ./models/cifar10/SE_ResNext29_best_cifar10.hdf5\n",
      "Epoch 2/200\n",
      "375/375 [==============================] - 68s 182ms/step - loss: 1.2146 - acc: 0.5646 - val_loss: 1.3263 - val_acc: 0.5380\n",
      "\n",
      "Epoch 00002: val_acc improved from 0.49700 to 0.53800, saving model to ./models/cifar10/SE_ResNext29_best_cifar10.hdf5\n",
      "Epoch 3/200\n",
      "375/375 [==============================] - 68s 182ms/step - loss: 1.0344 - acc: 0.6314 - val_loss: 1.1554 - val_acc: 0.5975\n",
      "\n",
      "Epoch 00003: val_acc improved from 0.53800 to 0.59750, saving model to ./models/cifar10/SE_ResNext29_best_cifar10.hdf5\n",
      "Epoch 4/200\n",
      "375/375 [==============================] - 68s 182ms/step - loss: 0.9282 - acc: 0.6723 - val_loss: 1.3801 - val_acc: 0.5655\n",
      "\n",
      "Epoch 00004: val_acc did not improve from 0.59750\n",
      "Epoch 5/200\n",
      "375/375 [==============================] - 68s 182ms/step - loss: 0.8058 - acc: 0.7146 - val_loss: 1.0703 - val_acc: 0.6310\n",
      "\n",
      "Epoch 00005: val_acc improved from 0.59750 to 0.63100, saving model to ./models/cifar10/SE_ResNext29_best_cifar10.hdf5\n",
      "Epoch 6/200\n",
      "375/375 [==============================] - 69s 182ms/step - loss: 0.7440 - acc: 0.7382 - val_loss: 0.9366 - val_acc: 0.6865\n",
      "\n",
      "Epoch 00006: val_acc improved from 0.63100 to 0.68650, saving model to ./models/cifar10/SE_ResNext29_best_cifar10.hdf5\n",
      "Epoch 7/200\n",
      "375/375 [==============================] - 69s 183ms/step - loss: 0.6702 - acc: 0.7656 - val_loss: 1.0623 - val_acc: 0.6515\n",
      "\n",
      "Epoch 00007: val_acc did not improve from 0.68650\n",
      "Epoch 8/200\n",
      "375/375 [==============================] - 68s 182ms/step - loss: 0.6205 - acc: 0.7841 - val_loss: 0.8071 - val_acc: 0.7100\n",
      "\n",
      "Epoch 00008: val_acc improved from 0.68650 to 0.71000, saving model to ./models/cifar10/SE_ResNext29_best_cifar10.hdf5\n",
      "Epoch 9/200\n",
      "375/375 [==============================] - 68s 182ms/step - loss: 0.5836 - acc: 0.7953 - val_loss: 0.9606 - val_acc: 0.7010\n",
      "\n",
      "Epoch 00009: val_acc did not improve from 0.71000\n",
      "Epoch 10/200\n",
      "375/375 [==============================] - 68s 182ms/step - loss: 0.5338 - acc: 0.8119 - val_loss: 0.7364 - val_acc: 0.7435\n",
      "\n",
      "Epoch 00010: val_acc improved from 0.71000 to 0.74350, saving model to ./models/cifar10/SE_ResNext29_best_cifar10.hdf5\n",
      "Epoch 11/200\n",
      "375/375 [==============================] - 68s 181ms/step - loss: 0.5012 - acc: 0.8224 - val_loss: 0.8323 - val_acc: 0.7395\n",
      "\n",
      "Epoch 00011: val_acc did not improve from 0.74350\n",
      "Epoch 12/200\n",
      "375/375 [==============================] - 68s 181ms/step - loss: 0.4727 - acc: 0.8315 - val_loss: 0.9347 - val_acc: 0.7095\n",
      "\n",
      "Epoch 00012: val_acc did not improve from 0.74350\n",
      "Epoch 13/200\n",
      "375/375 [==============================] - 68s 181ms/step - loss: 0.4421 - acc: 0.8437 - val_loss: 0.7834 - val_acc: 0.7345\n",
      "\n",
      "Epoch 00013: val_acc did not improve from 0.74350\n",
      "Epoch 14/200\n",
      "375/375 [==============================] - 68s 181ms/step - loss: 0.4105 - acc: 0.8573 - val_loss: 0.7358 - val_acc: 0.7560\n",
      "\n",
      "Epoch 00014: val_acc improved from 0.74350 to 0.75600, saving model to ./models/cifar10/SE_ResNext29_best_cifar10.hdf5\n",
      "Epoch 15/200\n",
      "375/375 [==============================] - 68s 182ms/step - loss: 0.3900 - acc: 0.8616 - val_loss: 0.7081 - val_acc: 0.7740\n",
      "\n",
      "Epoch 00015: val_acc improved from 0.75600 to 0.77400, saving model to ./models/cifar10/SE_ResNext29_best_cifar10.hdf5\n",
      "Epoch 16/200\n",
      "375/375 [==============================] - 68s 181ms/step - loss: 0.3701 - acc: 0.8733 - val_loss: 0.8757 - val_acc: 0.7410\n",
      "\n",
      "Epoch 00016: val_acc did not improve from 0.77400\n",
      "Epoch 17/200\n",
      "375/375 [==============================] - 68s 181ms/step - loss: 0.3419 - acc: 0.8796 - val_loss: 0.6210 - val_acc: 0.8040\n",
      "\n",
      "Epoch 00017: val_acc improved from 0.77400 to 0.80400, saving model to ./models/cifar10/SE_ResNext29_best_cifar10.hdf5\n",
      "Epoch 18/200\n",
      "375/375 [==============================] - 68s 181ms/step - loss: 0.3366 - acc: 0.8804 - val_loss: 0.6287 - val_acc: 0.7945\n",
      "\n",
      "Epoch 00018: val_acc did not improve from 0.80400\n",
      "Epoch 19/200\n",
      "375/375 [==============================] - 68s 180ms/step - loss: 0.3138 - acc: 0.8904 - val_loss: 0.9444 - val_acc: 0.7445\n",
      "\n",
      "Epoch 00019: val_acc did not improve from 0.80400\n",
      "Epoch 20/200\n",
      "375/375 [==============================] - 68s 181ms/step - loss: 0.3031 - acc: 0.8941 - val_loss: 0.7457 - val_acc: 0.7680\n",
      "\n",
      "Epoch 00020: val_acc did not improve from 0.80400\n",
      "Epoch 21/200\n",
      "375/375 [==============================] - 68s 181ms/step - loss: 0.2735 - acc: 0.9035 - val_loss: 0.7298 - val_acc: 0.7790\n",
      "\n",
      "Epoch 00021: val_acc did not improve from 0.80400\n",
      "Epoch 22/200\n",
      "375/375 [==============================] - 68s 181ms/step - loss: 0.2571 - acc: 0.9104 - val_loss: 0.6983 - val_acc: 0.7985\n",
      "\n",
      "Epoch 00022: ReduceLROnPlateau reducing learning rate to 0.001.\n",
      "\n",
      "Epoch 00022: val_acc did not improve from 0.80400\n",
      "Epoch 23/200\n",
      "375/375 [==============================] - 68s 181ms/step - loss: 0.2175 - acc: 0.9251 - val_loss: 0.4856 - val_acc: 0.8490\n",
      "\n",
      "Epoch 00023: val_acc improved from 0.80400 to 0.84900, saving model to ./models/cifar10/SE_ResNext29_best_cifar10.hdf5\n",
      "Epoch 24/200\n",
      "375/375 [==============================] - 68s 181ms/step - loss: 0.1640 - acc: 0.9450 - val_loss: 0.4854 - val_acc: 0.8515\n",
      "\n",
      "Epoch 00024: val_acc improved from 0.84900 to 0.85150, saving model to ./models/cifar10/SE_ResNext29_best_cifar10.hdf5\n",
      "Epoch 25/200\n",
      "375/375 [==============================] - 68s 181ms/step - loss: 0.1439 - acc: 0.9530 - val_loss: 0.4902 - val_acc: 0.8490\n",
      "\n",
      "Epoch 00025: val_acc did not improve from 0.85150\n",
      "Epoch 26/200\n",
      "375/375 [==============================] - 68s 181ms/step - loss: 0.1357 - acc: 0.9560 - val_loss: 0.4903 - val_acc: 0.8515\n",
      "\n",
      "Epoch 00026: val_acc did not improve from 0.85150\n",
      "Epoch 27/200\n",
      "375/375 [==============================] - 68s 181ms/step - loss: 0.1277 - acc: 0.9567 - val_loss: 0.4858 - val_acc: 0.8530\n",
      "\n",
      "Epoch 00027: val_acc improved from 0.85150 to 0.85300, saving model to ./models/cifar10/SE_ResNext29_best_cifar10.hdf5\n",
      "Epoch 28/200\n",
      "375/375 [==============================] - 68s 181ms/step - loss: 0.1234 - acc: 0.9605 - val_loss: 0.4933 - val_acc: 0.8510\n",
      "\n",
      "Epoch 00028: val_acc did not improve from 0.85300\n",
      "Epoch 29/200\n",
      "375/375 [==============================] - 68s 181ms/step - loss: 0.1163 - acc: 0.9628 - val_loss: 0.4916 - val_acc: 0.8535\n",
      "\n",
      "Epoch 00029: val_acc improved from 0.85300 to 0.85350, saving model to ./models/cifar10/SE_ResNext29_best_cifar10.hdf5\n",
      "Epoch 30/200\n",
      "375/375 [==============================] - 68s 181ms/step - loss: 0.1119 - acc: 0.9630 - val_loss: 0.4902 - val_acc: 0.8590\n",
      "\n",
      "Epoch 00030: val_acc improved from 0.85350 to 0.85900, saving model to ./models/cifar10/SE_ResNext29_best_cifar10.hdf5\n",
      "Epoch 31/200\n",
      "375/375 [==============================] - 68s 181ms/step - loss: 0.1107 - acc: 0.9645 - val_loss: 0.5015 - val_acc: 0.8530\n",
      "\n",
      "Epoch 00031: val_acc did not improve from 0.85900\n",
      "Epoch 32/200\n",
      "375/375 [==============================] - 68s 180ms/step - loss: 0.1038 - acc: 0.9668 - val_loss: 0.4935 - val_acc: 0.8590\n",
      "\n",
      "Epoch 00032: val_acc did not improve from 0.85900\n",
      "Epoch 33/200\n",
      "375/375 [==============================] - 68s 180ms/step - loss: 0.1012 - acc: 0.9681 - val_loss: 0.5064 - val_acc: 0.8590\n",
      "\n",
      "Epoch 00033: val_acc did not improve from 0.85900\n",
      "Epoch 34/200\n",
      "375/375 [==============================] - 68s 181ms/step - loss: 0.0965 - acc: 0.9699 - val_loss: 0.5109 - val_acc: 0.8610\n",
      "\n",
      "Epoch 00034: val_acc improved from 0.85900 to 0.86100, saving model to ./models/cifar10/SE_ResNext29_best_cifar10.hdf5\n",
      "Epoch 35/200\n",
      "375/375 [==============================] - 68s 180ms/step - loss: 0.0940 - acc: 0.9688 - val_loss: 0.5065 - val_acc: 0.8545\n",
      "\n",
      "Epoch 00035: val_acc did not improve from 0.86100\n",
      "Epoch 36/200\n",
      "375/375 [==============================] - 68s 181ms/step - loss: 0.0931 - acc: 0.9701 - val_loss: 0.5209 - val_acc: 0.8555\n",
      "\n",
      "Epoch 00036: val_acc did not improve from 0.86100\n",
      "Epoch 37/200\n",
      "375/375 [==============================] - 68s 181ms/step - loss: 0.0886 - acc: 0.9708 - val_loss: 0.5214 - val_acc: 0.8490\n",
      "\n",
      "Epoch 00037: val_acc did not improve from 0.86100\n",
      "Epoch 38/200\n",
      "375/375 [==============================] - 68s 181ms/step - loss: 0.0817 - acc: 0.9747 - val_loss: 0.5219 - val_acc: 0.8545\n",
      "\n",
      "Epoch 00038: val_acc did not improve from 0.86100\n",
      "Epoch 39/200\n",
      "375/375 [==============================] - 68s 181ms/step - loss: 0.0822 - acc: 0.9737 - val_loss: 0.5212 - val_acc: 0.8550\n",
      "\n",
      "Epoch 00039: ReduceLROnPlateau reducing learning rate to 0.001.\n",
      "\n",
      "Epoch 00039: val_acc did not improve from 0.86100\n",
      "Epoch 40/200\n",
      "375/375 [==============================] - 68s 181ms/step - loss: 0.0831 - acc: 0.9738 - val_loss: 0.5322 - val_acc: 0.8510\n",
      "\n",
      "Epoch 00040: val_acc did not improve from 0.86100\n",
      "Epoch 41/200\n",
      "375/375 [==============================] - 68s 181ms/step - loss: 0.0767 - acc: 0.9759 - val_loss: 0.5314 - val_acc: 0.8530\n",
      "\n",
      "Epoch 00041: val_acc did not improve from 0.86100\n",
      "Epoch 42/200\n",
      "375/375 [==============================] - 68s 181ms/step - loss: 0.0770 - acc: 0.9766 - val_loss: 0.5397 - val_acc: 0.8560\n",
      "\n",
      "Epoch 00042: val_acc did not improve from 0.86100\n",
      "Epoch 43/200\n",
      "375/375 [==============================] - 68s 180ms/step - loss: 0.0771 - acc: 0.9744 - val_loss: 0.5489 - val_acc: 0.8545\n",
      "\n",
      "Epoch 00043: val_acc did not improve from 0.86100\n",
      "Epoch 44/200\n",
      "375/375 [==============================] - 68s 181ms/step - loss: 0.0712 - acc: 0.9768 - val_loss: 0.5227 - val_acc: 0.8585\n",
      "\n",
      "Epoch 00044: ReduceLROnPlateau reducing learning rate to 0.001.\n",
      "\n",
      "Epoch 00044: val_acc did not improve from 0.86100\n",
      "Epoch 45/200\n",
      "375/375 [==============================] - 68s 181ms/step - loss: 0.0712 - acc: 0.9776 - val_loss: 0.5482 - val_acc: 0.8495\n",
      "\n",
      "Epoch 00045: val_acc did not improve from 0.86100\n",
      "Epoch 46/200\n",
      "375/375 [==============================] - 68s 181ms/step - loss: 0.0680 - acc: 0.9788 - val_loss: 0.5511 - val_acc: 0.8485\n",
      "\n",
      "Epoch 00046: val_acc did not improve from 0.86100\n",
      "Epoch 47/200\n",
      "375/375 [==============================] - 68s 181ms/step - loss: 0.0656 - acc: 0.9793 - val_loss: 0.5476 - val_acc: 0.8480\n",
      "\n",
      "Epoch 00047: val_acc did not improve from 0.86100\n",
      "Epoch 48/200\n",
      "375/375 [==============================] - 68s 180ms/step - loss: 0.0658 - acc: 0.9794 - val_loss: 0.5548 - val_acc: 0.8535\n",
      "\n",
      "Epoch 00048: val_acc did not improve from 0.86100\n",
      "Epoch 49/200\n",
      "375/375 [==============================] - 68s 180ms/step - loss: 0.0630 - acc: 0.9807 - val_loss: 0.5545 - val_acc: 0.8560\n",
      "\n",
      "Epoch 00049: ReduceLROnPlateau reducing learning rate to 0.001.\n",
      "\n",
      "Epoch 00049: val_acc did not improve from 0.86100\n"
     ]
    }
   ],
   "source": [
    "from tensorflow.keras.models import Model\n",
    "from tensorflow.keras.preprocessing.image import ImageDataGenerator\n",
    "from tensorflow.keras.callbacks import LearningRateScheduler,ModelCheckpoint,ReduceLROnPlateau\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "path='./models/cifar10/SE_ResNext29_best_cifar10.hdf5'\n",
    "train_datagen = ImageDataGenerator(\n",
    "        horizontal_flip=True,\n",
    "        width_shift_range=4, #+-8 pixel shift max\n",
    "        height_shift_range=4)#+-8 pixel shift max\n",
    "\n",
    "\n",
    "reduce_lr = ReduceLROnPlateau(monitor='val_acc', mode='max',factor=0.1,\n",
    "                              patience=5, min_lr=0.001, verbose=1)\n",
    "\n",
    "#Early stopping\n",
    "es = tf.keras.callbacks.EarlyStopping(monitor='val_acc', mode='max', patience=15)\n",
    "# checkpoints\n",
    "checkpoint = ModelCheckpoint(path, monitor='val_acc', verbose=1, save_best_only=True, mode='max')\n",
    "SEResNeXt29.compile(optimizer=tf.keras.optimizers.SGD(learning_rate=0.01,momentum=0.9),loss='categorical_crossentropy', metrics = ['acc'])\n",
    "\n",
    "\n",
    "train_generator = train_datagen.flow(X_train, y_train, batch_size=128)\n",
    "step_size_train=train_generator.n//train_generator.batch_size\n",
    "\n",
    "history=SEResNeXt29.fit(train_generator,\n",
    "                    steps_per_epoch=step_size_train,\n",
    "                    epochs=200,\n",
    "                    validation_data=(X_val,y_val),\n",
    "                    callbacks=[reduce_lr,es,checkpoint])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "id": "1867acf6-766b-454d-85f4-0415dbb2ac88",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 576x576 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "from utils.train_CIFAR_all import plot_model\n",
    "plot_model(history)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "id": "b47f0d4b-8477-4068-8ac1-593c3dd837c5",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[0.4397708475589752, 0.5788958072662354, 0.6378333568572998, 0.6726250052452087, 0.7133958339691162, 0.7384583353996277, 0.7637500166893005, 0.7783750295639038, 0.7921249866485596, 0.8090416789054871, 0.8205000162124634, 0.8323958516120911, 0.8423958420753479, 0.8531458377838135, 0.8594375252723694, 0.8684791922569275, 0.8770624995231628, 0.8808541893959045, 0.8893958330154419, 0.8948333263397217, 0.8998333215713501, 0.9067916870117188, 0.9344375133514404, 0.9481250047683716, 0.953083336353302, 0.9549375176429749, 0.9568750262260437, 0.9599166512489319, 0.9616458415985107, 0.9634583592414856, 0.9651041626930237, 0.9670416712760925, 0.9673749804496765, 0.9693750143051147, 0.968874990940094, 0.9698749780654907, 0.9714375138282776, 0.9737916588783264, 0.9736250042915344, 0.9735833406448364, 0.9751458168029785, 0.9758541584014893, 0.9751458168029785, 0.9764999747276306, 0.9781249761581421, 0.9779999852180481, 0.9790208339691162, 0.9788333177566528, 0.9800208210945129]\n",
      "[1.58455228805542, 1.1804163455963135, 1.018962025642395, 0.9273545742034912, 0.810274600982666, 0.7397942543029785, 0.6727392673492432, 0.6324736475944519, 0.5918411612510681, 0.5439109802246094, 0.5126926898956299, 0.47665882110595703, 0.4476388692855835, 0.4197205901145935, 0.39847368001937866, 0.37778422236442566, 0.34811487793922424, 0.3401123881340027, 0.3152509331703186, 0.30345046520233154, 0.2816046476364136, 0.2650785744190216, 0.1924656331539154, 0.15701346099376678, 0.1423623114824295, 0.1364123672246933, 0.1281486451625824, 0.12244122475385666, 0.1195472851395607, 0.11295976489782333, 0.1082097589969635, 0.10253369808197021, 0.1012948602437973, 0.0965530127286911, 0.09465303272008896, 0.09303110837936401, 0.08762841671705246, 0.08326134830713272, 0.08225119858980179, 0.08357013761997223, 0.07801725715398788, 0.0768180564045906, 0.0767098069190979, 0.07276783883571625, 0.06997157633304596, 0.06985743343830109, 0.0660567432641983, 0.0664624348282814, 0.06388210505247116]\n",
      "[0.4970000088214874, 0.5379999876022339, 0.5975000262260437, 0.565500020980835, 0.6309999823570251, 0.6865000128746033, 0.6514999866485596, 0.7099999785423279, 0.7009999752044678, 0.7434999942779541, 0.7394999861717224, 0.7095000147819519, 0.734499990940094, 0.7559999823570251, 0.7739999890327454, 0.7409999966621399, 0.8040000200271606, 0.7944999933242798, 0.7444999814033508, 0.7680000066757202, 0.7789999842643738, 0.7985000014305115, 0.8489999771118164, 0.8514999747276306, 0.8489999771118164, 0.8514999747276306, 0.8529999852180481, 0.8510000109672546, 0.8535000085830688, 0.859000027179718, 0.8529999852180481, 0.859000027179718, 0.859000027179718, 0.8610000014305115, 0.8544999957084656, 0.8554999828338623, 0.8489999771118164, 0.8544999957084656, 0.8550000190734863, 0.8510000109672546, 0.8529999852180481, 0.8560000061988831, 0.8544999957084656, 0.8585000038146973, 0.8495000004768372, 0.8485000133514404, 0.8479999899864197, 0.8535000085830688, 0.8560000061988831]\n",
      "[1.4102625846862793, 1.3263078927993774, 1.1553984880447388, 1.380059003829956, 1.0703016519546509, 0.9366141557693481, 1.0622822046279907, 0.8070918917655945, 0.9605752825737, 0.7364020943641663, 0.8322935104370117, 0.9346789121627808, 0.7833690047264099, 0.7358474731445312, 0.7080677151679993, 0.8757030963897705, 0.6209886074066162, 0.6286855936050415, 0.9444307684898376, 0.7456822395324707, 0.7298475503921509, 0.6983386874198914, 0.4855524003505707, 0.4853965938091278, 0.4902157187461853, 0.4903382956981659, 0.48581215739250183, 0.49327409267425537, 0.491594523191452, 0.4901910424232483, 0.5014814734458923, 0.4935123920440674, 0.5063860416412354, 0.5109213590621948, 0.5065250396728516, 0.5209395885467529, 0.5214101672172546, 0.5219017863273621, 0.5212174654006958, 0.5321537256240845, 0.5314016342163086, 0.5396841764450073, 0.5488673448562622, 0.5226861834526062, 0.5482343435287476, 0.5510746240615845, 0.5475994348526001, 0.5547808408737183, 0.554492712020874]\n"
     ]
    }
   ],
   "source": [
    "print(history.history['acc'])\n",
    "print(history.history['loss'])\n",
    "print(history.history['val_acc'])\n",
    "print(history.history['val_loss'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "id": "73e9aea5-c9bb-4b78-87cf-f620dbe5a23d",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "313/313 - 14s - loss: 0.6006 - acc: 0.8309\n",
      "SE_custom_ResNext29_8x64d\n",
      "model accuracy: 83.09%\n",
      "top 3 accuracy: 96.60000205039978\n",
      "top 5 accuracy: 99.14000034332275\n"
     ]
    }
   ],
   "source": [
    "# reload from saved weights:\n",
    "model=SE_ResNeXt29(input_shape=(32,32,3),n_classes=10,ratio=16,name='custom_ResNeXt29_SE')\n",
    "model.load_weights(path)\n",
    "\n",
    "if not os.path.exists('./full_models/cifar10'):\n",
    "    os.makedirs('./full_models/cifar10')\n",
    "model.save('./full_models/cifar10/SE_ResNext29_best.h5')\n",
    "\n",
    "\n",
    "score(model,X_test,y_test,top_n=[3,5])"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.9"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}