Deploying a ML model on AWS SageMaker

Udacity Machine Learning Engineer Nanodegree Program | Machine Learning in Production Project

This repository contains my project for the Machine Learning in Production module of the Machine Learning Engineer Nanodegree from Udacity.

Boiler code and utility functions were provided directly by Udacity and are not of my work. Udacity Github repository

AWS API Gateway has changed a bit since Udacity's code was released, but the general idea stands.

One thing not mentioned in the notebook is to set up CORS coorectly for the created API.

The report.html file is the exported notebook in HTML.

Overview

This project aims to deploy a machine learning model on AWS.

The model will be an LTSM neural network used to classify movie reviews in two categories: positive or negative.

Both model training and deployment is done on AWS SageMaker, and we used PyTorch as NN framework.

Then a simple web page allows to input a movie review and have it classified by the deployed model thanks to AWS API Gateway and Lambda Services:

• API Gateway is used solely for forwarding the request to the defined lambda
• Lambda service is used to handle the HTTP request and response, and invoke our model.

Files

.
├── assets                  # Images
├── train                   # Python scripts to define and train the PyTorch model
├── serve                   # Python scripts to load the trained PyTorch model and handle data (de)serialization
├── index.html              # Simple UI to input a review to classify and retrieve model
├── README.md
├── report.html             # Exported notebook to HTML
└── SageMaker Project.ipynb # Jupyter notebook

Demo

I might leave the model online for some times depending on AWS free tier...

Sentiment analysis webapp

HTML report

Jupyter Notebook HTML