This is an ML Project

Instructions:

• clone repository
• Run make install to install dependencies
• Run make test to test end-to-end workflow
• Optional:
• Run python run.py to execute workflow directly
• Run uvicorn inference:app --reload --host 0.0.0.0 --port 8001 to serve model on port 8001
• After serving model (previous command), run python ml_stuff/models/test_api to query endpoint

Overview

Continual helps ML teams bring software engineering practices to ML deployments. This project is a simple example.

• Start with a basic project structure:

• Code that trains a model (train.py)

• Code that tests our model (test_train.py)

• Script that builds and tests our code (Makefile)

• CI job to automate build & test on merge (GH Action - main.yml)

• The second iteration adds inferencing and Continuous Deployment

• CD job deploys REST API to AWS ECS on pull request

• REST API for model inferencing with FastAPI (inference.py)

• Code that tests our API (test_api.py)

• Dockerfile to containerize our web service

• CD via AWS ECS (GH Action --> ECS Fargate)

• Right now - this deployment requires manual review by me

• To Do:

• Add multiple environments

• Add mocks for tests

• Add orchestration

• Sagemaker, AzureML, Vertex