October 12, 2022
Machine learning is experimental in nature. It’s more like research in a lab than it…
When we talk about the challenges of building ML-powered applications, we tend to focus on problems related to the quality of a model’s predictions — things like data drift, changes in model architectures, or inference latency.
But there’s an often overlooked challenge that plays a vital role in the success or failure of these applications— the process of integrating a model into existing software.
ML is a highly iterative discipline. In the process of developing a model, teams often make many changes to their codebase and pipelines. Coupling an ML codebase to an application’s dependencies, unit tests, and CI/CD pipelines will significantly reduce the velocity with which ML teams can deliver on a solution—this is because each change will almost assuredly require running these downstream dependencies before a merge can be approved.
It’s with these unique challenges in mind that we’re thrilled to share our latest integration with GitLab, a powerful, pipeline-based DevOps platform that helps teams better manage their CI/CD workflows.
With this integration between the GitLab DevOps platform and Comet, we can keep workflows between ML and Engineering teams separate, while enabling cross-team collaboration—by preserving the visibility and audit-ability of the entire model development process across teams.
Using the combined power of Comet and GitLab, ML and application teams can implement a robust, seamless process where discussions, code reviews, and model performance metrics get automatically published and tracked within Merge Requests on the GitLab DevOps platform.
With this process in place, teams can increasing the velocity and opportunity for collaboration between data scientists and software engineers for machine learning workflows.
To get started with this powerful integration, we’ve compiled a number of resources:
In this video, GitLab’s Technical Marketing Manager, William Arias, walks through the process of building the required pipelines and workflows for a simple demo project