Skore let's you "Own Your Data Science." It provides a user-friendly interface to track and visualize your modeling results, and perform evaluation of your machine learning models with scikit-learn.
skops is a secure persistence format for scikit-learn compatible models.
It does not allow to persist everything for several reasons:
some models need specific support from skops, for example, scikeras models that are tensorflow/scikit-learn hybrid models.
some properties in scikit-learn estimators are untrusted because they are a security risk.
not all scikit-learn properties are added to the white list because they are yet to be discovered (see #347)
We want to promote the use of skops and help develop this project, so it is not a priority to support alternative serialization formats for scikit-learn-compatible estimators.
Right, now if a user saves a model in skore, they might not be able to load it back because of untrusted properties. It is required to tell skops what properties are trusted at load time.
We can alleviate this in two steps:
Fix #400
Provide an API in skore to manage skops trusted properties.
Fix #347
Present a nice UX where the user could see the warnings, and opts to save the trusted properties. Then, it would be our job to prevent tampering with trusted properties, and use the list at load time.
Following a discussion with @adrinjalali
skopsis a secure persistence format for scikit-learn compatible models. It does not allow to persist everything for several reasons:We want to promote the use of
skopsand help develop this project, so it is not a priority to support alternative serialization formats for scikit-learn-compatible estimators.Right, now if a user saves a model in
skore, they might not be able to load it back because of untrusted properties. It is required to tellskopswhat properties are trusted at load time.We can alleviate this in two steps:
skoreto manageskopstrusted properties.We must implement (1) before the public alpha.