Toy project, that implements passive retraining for a batch prediction regression use case.

The project is not concerned with deployment or the development of the ML models. It uses the Boston housing data and can be run locally.

• Hydra for configuration
• Weights and Biases for experiment tracking and artifact versioning
• MLFlow for model packaging
• Evidently AI for drift detection

• Weights and Biases account, which is free for personal use: https://wandb.ai/home. When you run one of the pipelines, you will be prompted for an API key, which can be found under settings.
• Conda

conda env create --file conda.yml 

pip install -e .

make train_pipeline

This will run a training pipeline that will train a model, test it and potentially promote it to production status (by tagging the model arrtifact with a prod tag.

make inference_pipeline

This will run an inference pipeline that will use the prod model to make predictions on new data (just a sample from the Boston housing data). A very simplistic drift can be configures in the main Hydra configuration.

make drift_detection

This will run drift detection, that compares the data used to make the latest predictions with the data used to train the latest prod model.

make sweep_random_forest

This will prepare a Weights and Biases hyperparameter sweep. You will be prompted in the terminal on how to actually run the sweep agent.