Fix add features

[guolinke]

fix #2744 and fix #2603

@StrikerRUS could you help for the test cases?

• test cases
• merge data sources at Python side

[StrikerRUS]

@guolinke

could you help for the test cases?

Sure, but #2594 removes test_add_features_* tests completely. I think we should merge it firstly.

[guolinke]

@StrikerRUS
sorry for missing that. I will fix it when I have time.

[guolinke]

ping @StrikerRUS

[guolinke]

@StrikerRUS do you think this is ready?

[StrikerRUS]

do you think this is ready?

No, I don't think so. Python package still lacks tests and logic for updating feature names and category fields of Dataset class (#2754 (comment)) (maybe something more). I'm sorry, I really don't have a lot of time right now to propose a thoughtful code update. PR reviews looks more urgent and takes much time, unfortunately.
I believe that this PR should not be blocking us from 3.0 release as there are no breaking changes here, only bug fixes that we can release in a next minor release.

[guolinke]

@StrikerRUS no problem, we can skip this for now.

[StrikerRUS]

Also fix #3221.

[StrikerRUS]

@guolinke

Speaking about categorical features, it looks like it is not trivial to set them back at the Python side after calling AddFeaturesFrom(). In other dataset they can be provided by names, by indices, or even have auto value with pandas dataframe. The same cases are valid for self dataset. I don't know how to merge all possible combinations properly. Moreover, after cpp side we can have changed feature names (in case of duplicated names). I think the best option for now is just to warn users about this.

TBH, I don't want to hold this PR more...

[guolinke]

@StrikerRUS Thanks very much.
I think this is a rare use case, so no problem for the incomplete function.
As long as the numerical features could be merged, I think it is okay.

[guolinke]

@StrikerRUS I cannot approve my own PR, you can approve and merge it if you think it is ready

[StrikerRUS]

@guolinke Thank you! I think it's ready.

[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.