Reproducible ML Experimentation with Rubicon-ML

Explore reproducible machine learning experimentation with 'rubicon-ml' in this 32-minute conference talk from All Things Open 2022. Learn how Capital One's open-source Python library captures and stores model training and execution information, ensuring full reproducibility and audit-ability for developers and stakeholders. Discover how to seamlessly incorporate 'rubicon-ml' into existing ML workflows, leveraging its compatibility with popular tools like git, Scikit-learn, Dask, and Plotly. Gain insights into logging model metadata locally, using various backends, sharing and comparing experiments, and visualizing results through interactive dashboards. Understand the importance of tracking iterations across teams and see how 'rubicon-ml' addresses these challenges in Capital One's ML processes.

Intro
Building a model is an iterative process
Tracking this iteration is important to developers and stakeholders
Tracking this iteration across a team can be difficult
We built and open sourced rubicon-ml to help!
Logging locally as a developer leveraging Scikit-learn
Other logging backends with £aspec
Sharing and comparing experiments with intake.
Visualizing experiments with Dash & Plotly
Integrating rubicon-ml into ML workflows at Capital One