Return training loss from LogisticRegression

[rachrapp]

Describe the workflow you want to enable

Currently there seems to be no way to retrieve the loss (or change in loss) when training a logistic regression model (sklearn.linear_model.LogisticRegression). If one changes the verbosity then this can be seen from the terminal, but also cannot be retrieved from stdout. This makes it unfeasible to actually plot training curves from logistic regression models.

Describe your proposed solution

Add train_loss as an attribute, same as e.g. classes_ or n_features_in_.

Describe alternatives you've considered, if relevant

No response

Additional context

No response

[asingh9530]

Hi @rachrapp could you check out this link here. Seems like you need to define a custom wrapper around it.

[rachrapp]

Thanks @Abhinavfreecodecamp. While this workaround does work for the SGDClassifier class, it unfortunately does not work for LogisticRegression.

[jeremiedbb]

This is the kind of feature that will be possible when we have a callback API (#22000), which is still work in progress.