Callback API plan

[jeremiedbb]

The goal of this issue is to track the steps of the implementation of a callback API in scikit-learn.

This is being developed in the callbacks feature branch. The first PR to this branch is #27663 which implements the base infrastructure for the callbacks and a first callback (progress bars). Subsequent PRs will add more callbacks, mode doc, more tests, adapt estimators to support callbacks, ...

• Base infrastructure FEA Callbacks base infrastructure + progress bars #27663

Callbacks

• Progress bars FEA Callbacks base infrastructure + progress bars #27663
• Monitoring
• EarlyStopping
• Snapshots
• Verbose / Logging

Doc

• examples
• how to write a custom callback

Adapt estimators
full list incoming

[lorentzenchr]

As callbacks affect all estimators and add to the API, a SLEP would not be wrong.

[jondo]

@glemaitre meant that one such callback could be a gradient tape. I would currently need that (and might even be able to help implementing it).