sagemaker-custom-model

This repository is a collection of commonly used containers that can be deployed easily on sagemaker and addresses common issues like installing specific versions of python/R which become problematic while creating custom containers on docker

Structure of Repository

The following structure will be followed for subsequent modelling containers in the repositories

.
├── model-name                                                # catboost/lightgbm/xgboost etc
    ├── Readme.md                                             # Readme file on how to use test model on local Docker and then train on sagemaker
    ├── Dockerfile                                            # Dockerfile to create container
    ├── build_and_push.sh                                     # Build and push container
    ├── local_test                                            # Material to test local deployment of code
    |    ├── train_local.sh                                   # Train using container locally
    |    ├── serve_local.sh                                   # Serve model locally
    |    ├── predict.sh                                       # Predict model on local serving using payload
    |    ├── payload.csv                                      # payload for local testing
    |    └── test_dir                                           
    |            ├── input
    |            |   ├── config 
    |            |   |   └── hyperparameters.json             #Hyperparameters for local testing
    |            |   └── data
    |            |       └── training
    |            |           └──  train.csv                   #Training data (1000 rows) for local testing
    |            └── model
    |                └── model object                         #Once trained locally model object will be saved here
    └── model_container
          ├── nginx.conf                                      #Setup server
          ├── predictor.py                                    #Prediction function
          ├── serve                                           #Implements the scoring service shell
          ├── train                                           #Training Code 
          └── wsgi.py                                         #Wrapper for gunicorn