Now you can easily find and organize your experiments with filtered views based on experiment metrics, metadata, and parameters

Machine learning teams often work with many, many models and their iterations (often in the hundreds and thousands 📈). We received overwhelming feedback from our users about needing improved discoverability for their machine learning experiments.

To enable our users to better sort through and discover their most interesting and highly performing experiments, our team implemented a feature called the Query Builder.

With the Query Builder, Comet.ml users are able to do more complex and effective experiment filtering so you only see relevant experiments.

A quick snapshot of the Comet.ml Query Builder

A Closer Look:

Let’s do a quick walkthrough of how the Query Builder can help you and your team. You can try the Query Builder today by signing up for a Comet.ml account!

Making a Query

With the query results, you can now easily see what your top-performing experiments are (ex. AUC score > 0.95) and sort even further within the subset of experiments.

The Query Builder in action!

💥 Note: all of our filters are AND operators for now

Saving Query

You can also save a set of filters as a Saved Query so that you can access the same subset of experiments again or see what new experiments match the filters you set.

💥Pro Tip: if you’re on a Teams Comet.ml account, your Saved Queries will also accessible to your teammates who are on the same project!

In your Query dropdown, you can see all your Saved Queries and also easily navigate back to your All Experiment view. In this sample project, I have three Saved Queries related to ROC scores and learning rates that my teammates can also access!

Combining Query + Group By

To kick it up another organizational notch, you can also combine the Query Builder with our new Group By feature. This allows you to group the filtered subset you already have (made with the Query Builder) in yet another level. Very useful for organizing cross-validation runs!

In this sample project, I’m grouping my filtered experiment list by the two different learning rates I ran my models with — much easier to see what the impact of learning rate on performance is!

Comet.ml’s Query Builder allows for exponentially improved experiment discovery, transparency, and organization.

If you’ve been looking for a better way to track, manage, and compare your machine learning experiments, try out Comet.ml today for free! Our inbox and Slack community are always open for feedback 🙌


Enjoyed this article? Here are some other articles you might enjoy:


About Comet.ml — Comet.ml is doing for ML what Github did for code. Our lightweight SDK enables data science teams to automatically track their datasets, code changes, experimentation history. This way, data scientists can easily reproduce their models and collaborate on model iteration amongst their team!

Posted by:Cecelia Shao

Product Lead @ Comet.ml Comet is doing for Machine Learning what GitHub did for software. We allow data science teams to automatically track their datasets, code changes, experimentation history and production models creating efficiency, transparency, and reproducibility. Learn more at www.comet.ml

Leave a Reply

Your email address will not be published. Required fields are marked *