rubicon-ml: Capital One’s open supply answer to standardize the mannequin improvement lifecycle


In an period of fixed innovation, there’s an growing want for steady iteration, which ends up in a extra advanced mannequin improvement lifecycle. Maintaining observe of all of the inputs and outputs together with options, metrics and artifacts for every mannequin model may be troublesome and, at occasions, tedious. 

In pursuit of simplifying this course of, Capital One created rubicon-ml, an open-source machine studying (ML) answer that may observe, visualize and share experiments with collaborators and reviewers. These capabilities can assist information scientists and technologists experiment, prepare and govern fashions designed to resolve advanced enterprise issues.

“Earlier than a mannequin is definitely pushed to manufacturing, ML specialists conduct hundreds of experiments with completely different enter parameters that end in numerous outputs,” mentioned Sri Ranganathan, director of ML engineering at Capital One and proprietor of rubicon-ml. “Rubicon tracks these experiments throughout the mannequin improvement lifecycle and may present the standing of code for any given parameter.” 

Ranganathan went on to clarify that rubicon-ml simplifies mannequin governance, auditability and reproducibility by explaining how numerous parameters affect the general output of a mannequin. ”This may be significantly helpful for an inner Mannequin Threat Workplace as they search to approve, validate and govern fashions throughout a corporation,” she mentioned.

rubicon-ml is simple to make use of and integrates immediately right into a consumer’s Python mannequin pipeline. It leverages present open supply tooling together with Scikit-learn for mannequin coaching; Sprint and Plotly for visualizations; and Consumption for sharing experimental outcomes. In line with Ranganathan, what differentiates rubicon-ml from different related instruments at present available on the market is the facility that it offers to the consumer to decide on the platform or file format.

“The open supply nature of this answer additionally units it aside from competing instruments,” mentioned Nureen D’Souza, director of the Open Supply Program Workplace at Capital One. With contributions from specialists throughout the ecosystem, open supply software program creates a high-quality product that grows even stronger over time.

In line with D’Souza, it’s vital for Capital One to provide again to the open supply group and work collectively to enhance the software program that everybody wants. By open sourcing our options, we are able to make a a lot greater affect than would have ever been potential in any other case. “Plus, we all know that open supply software program improvement creates higher high quality and safer code.”

As rubicon-ml continues to steadily develop, Ranganathan mentioned that Capital One is at all times trying to make enhancements to the answer. ”We’re planning to make new integrations with the most recent Python ML libraries. And we’re at all times on the lookout for new contributions to make the answer even stronger.”

D’Souza and Ranganathan will probably be talking about rubicon-ml and different new open supply options from Capital One on the All Issues Open convention later this 12 months. Within the meantime, go to rubicon-ml on GitHub to find out how the answer can standardize the mannequin improvement lifecycle at your group.

Similar Posts

Leave a Reply

Your email address will not be published.