Upgrade to Pro — share decks privately, control downloads, hide ads and more …

Software engineering practices applied to ML

Software engineering practices applied to ML

PyData London Meetup - April 2018 (https://www.meetup.com/PyData-London-Meetup/events/248973985/)

Stefano and Pavlos are engineers at HomeAway, and they'll talk about the benefits of applying good engineering practices when building ML-powered systems. The agenda will include:
- Feature engineering: does it have the same behaviour in training and prediction phases?
- CI/CD: are you checking for performance regressions in an automated fashion?
- Monitoring of your ML pipeline: are you sure your ML model works in production?

Avatar for Stefano Bonetti

Stefano Bonetti

April 03, 2018
Tweet

More Decks by Stefano Bonetti

Other Decks in Programming

Transcript

  1. def test_ml_models(hold_out_data_set): # For Baseline Model baseline_mae, baseline_rmse = _calculate_error_for_baseline_model(

    hold_out_data_set ) # For New Model new_model_mae, new_model_rmse = _calculate_error_for_new_model( hold_out_data_set ) assert new_model_mae < baseline_mae assert new_model_rmse < baseline_rmse