[ENH] Hyperparameter hits boundary warning

[samihamdan]

throw a warning when the any of the best hyperparameter is at the boundary

[fraimondo]

This can be implemented inside the run_cross_validation function by inspecting the models after fitting, no? Or we need to patch the *SearchCV objects to check?

[fraimondo]

Implementation:

After the internal call to cross_validate, check if we are doing any of the Grid/Random search and raise warning if the best_hyperparameters of each fold are touching the boundaries.

Caveat: we need to call cross_validate with return estimator = True for each fold, which will increase memory usage if models are too big.

We need to think of an alternative to turn this off, without adding a parameter to run_cross_validation.