Deploy and monitor a machine learning workflow for Image Classification

Introduction

In this project we implement a simple classification task: predict bicycle / motorcycleusing the CIFAR dataset. The main focus of the project is to deploy and monitor the machine learning workflow in AWS.

Steps

The full work can be seen in the notebook ML_Workflow.ipynb. The main steps are:

1. Load the CIFAR dataset, transform, and upload to S3.
2. Train a simple model using the SageMaker framework image-classification.
3. Deploy the model with monitoring and create an endpoint, while verifying we can make predictions.
4. Next, we leave the notebook and create 3 Lambda functions to make predictions (get image, classify, and warn on low confidence). See the lambdas folder.
5. We glue them together in a Step Function (see step_function folder).
6. Finally, we make some predictions and plot custom graphs to monitor the workflow.