mcleonte / mlops-stageflow

Streamlined switching between local and cloud development stages within a unified project filesystem.

MLOps StageFlow

Streamlined switching between local and cloud development stages within a unified project filesystem.

Description

This Machine Learning project is meant to showcase a streamlined transition between different development stages. It features an intuitive mechanism for toggling between environments, ensuring an optimised workflow for both a local environment, as well as a cloud-based environment on Google Cloud Platform.

Key Features

• Easy Environment Switching: Toggle ENV_MODE in the .env file to seamlessly switch between debug, dev, staging, and prod environments.
• centralized configuration: .env centralizes critical variables for automated resource naming, model configuration and versioning, and differentiated infrastructure specifications.
• simplified command structure: execute build, run, and test commands with ease across training and inference stages.

How to Use

• clone the repository

• configure the .env file according to your environment needs

• use the following commands to manage your project:

  training:

  • make training-build
  • make training-run
  • make training-test

  inference:

  • make inference-build
  • make inference-run
  • make inference-test

Upcoming Features

• preprocessors versioning
• training evaluation for staging and production stages using Vertex AI evaluation job (programatic implementation currently not supported)
• improved logs for debugging stage
• support for diferentiated training dataset for production stage