[python-package] support customizing Dataset creation in Booster.refit() (fixes #3038)

[TremaMiguel]

Context:
This PR aims to close #3038

Changes

• refit method accepts kwargs_for_dataset parameter to pass weight parameter to Dataset initialization inside refit().
• refit method accepts kwargs_for_predict parameter to pass original params to predict method.

Tests:
Based on @jtilly example on issue #3038 test that refit method returns a valid prediction value when passing kwargs_for_dataset or kwargs_for_predict.

[ghost]

All CLA requirements met.

[jameslamb]

Thanks for picking this up! I left some initial comments based on the format of what you're proposing, but I'm not the best person to comment on whether or not #3038 should be adopted at all.

For example, I don't know if the comment mentioned at #3038 (comment) means that init_score, weight, and group is still true.

Hopefully @guolinke @shiyu1994 or @StrikerRUS can comment on whether we should proceed with this feature.

[shiyu1994]

For example, I don't know if the comment mentioned at #3038 (comment) means that init_score, weight, and group is still true.

I think it is useful that we want to change the weights of data points when refit. If we see weight, init_score and group as properties of a dataset, then when refitting on a new dataset, it is very meaningful to allow new settings of these properties.

For example, a new ranking dataset can have totally different group.

[shiyu1994]

@TremaMiguel Thanks for working on this!

[jameslamb]

Thanks for the testing changes. Please see a few more suggestions I've provided.

[jameslamb]

Thanks very much. Please see some additional suggestions to make the tests a bit stronger.

[jameslamb]

Looks good to me, thanks very much!

I've edited the PR description so it will be a bit more informative when it's used as a bullet point in the release notes.

Before this is merged, I'd like another Python maintainer like @jmoralez or @StrikerRUS to review.

[StrikerRUS]

Thanks for working on this! Generally LGTM, just few minor fixes for the docstring.

[TremaMiguel]

@StrikerRUS @jameslamb @jmoralez is there any pending change to close this PR?

[StrikerRUS]

LGTM! Thank you so much for your contribution!

[github-actions]

This pull request has been automatically locked since there has not been any recent activity since it was closed. To start a new related discussion, open a new issue at https://github.com/microsoft/LightGBM/issues including a reference to this.