Alternative to MLFLow

[harrytruckit]

Hi,

I made a similar discussion post on Skrub and I'm wondering if this library is somewhat intending to compete with experiment tracking frameworks like MLFLow? So far I really like the simplicity of recording experiments using Skore, but with the main benefit of using MLFLow being the ease of deploying inference endpoints. Has anyone managed to combine using Skore and MLFlow well, and if so what is you "flow"?

Thanks!

[nbesson-probabl]

Hello @harrytruckit

Thank you for your interest in our library ! I'm Nicolas, the product manager for Skore

To quickly answer your question : this library is not somewhat intending to compete with experiment tracking frameworks like MLFLow, and it can be used as a complement to them.

The idea is to provide the missing data science components in those - MLflow included (advanced plotting, cross-validation experiment management, etc.).

However, we foresee providing tighter integration with MLflow at some point to offer features such as:

• Loading models from MLflow's artifact storage and computing reports
• Storing Skore reports and inner artifacts directly in MLflow storage
• Additional interoperability capabilities

Yet, please note that this library is part of a bigger Skore project.

The same way we provide this library as a complement to MLflow, we are working on providing an app (not under the same licence) with an UI dedicated to the exploration and presentation of Skore reports, with collaboration-oriented features.

We think there's a need for a complementary interface to MLFlow's one, especially when manipulating Skore reports.

But as for the lib, interoperability with tools as used as MLFlow will be among our top priorities.
You'll always be able to work with Skore during the experimentation phase, and still log your models in MLFlow or the registry of your choice at the end of it.

I hope this helps, feel free to reach out to us if you have any other question !