In an era of constant innovation, there is an increasing need for continuous iteration, which leads to a more complex model development lifecycle. Keeping track of all the inputs and outputs including features, metrics and artifacts for each model version can be difficult and, at times, tedious. 

In pursuit of simplifying this process, Capital One created rubicon-ml, an open-source machine learning (ML) solution that can track, visualize and share experiments with collaborators and reviewers. These capabilities can help data scientists and technologists experiment, train and govern models designed to solve complex business problems.

“Before a model is actually pushed to production, ML specialists conduct thousands of experiments with different input parameters that result in various outputs,” said Sri Ranganathan, director of ML engineering at Capital One and owner of rubicon-ml. “Rubicon tracks these experiments across the model development lifecycle and can provide the status of code for any given parameter.” 

Ranganathan went on to explain that rubicon-ml simplifies model governance, auditability and reproducibility by explaining how various parameters impact the overall output of a model. “This can be particularly useful for an internal Model Risk Office as they seek to approve, validate and govern models across an organization,” she said.

rubicon-ml is easy to use and integrates directly into a user’s Python model pipeline. It leverages existing open source tooling including Scikit-learn for model training; Dash and Plotly for visualizations; and Intake for sharing experimental results. According to Ranganathan, what differentiates rubicon-ml from other similar tools currently on the market is the power that it gives to the user to choose the platform or file format.

“The open source nature of this solution also sets it apart from competing tools,” said Nureen D’Souza, director of the Open Source Program Office at Capital One. With contributions from experts across the ecosystem, open source software creates a high-quality product that grows even stronger over time.

According to D’Souza, it’s important for Capital One to give back to the open source community and work together to improve the software that everyone needs. By open sourcing our solutions, we can make a much bigger impact than would have ever been possible otherwise. “Plus, we know that open source software development creates better quality and more secure code.”

As rubicon-ml continues to steadily grow, Ranganathan said that Capital One is always looking to make improvements to the solution. “We’re planning to make new integrations with the latest Python ML libraries. And we’re always looking for new contributions to make the solution even stronger.”

D’Souza and Ranganathan will be speaking about rubicon-ml and other new open source solutions from Capital One at the All Things Open conference later this year. In the meantime, visit rubicon-ml on GitHub to learn how the solution can standardize the model development lifecycle at your organization.