Fix some paramater hints when loading from binary file

[hzy46]

#4657 describes ignore_column will cause an error when loading from binary file. I think here're two related issues in the code:

1. SetHeader should only be called when loading from text file. Now, it is also called when loading the data from binary file. This is why error mentioned in issue 4657 happens.

LightGBM/src/io/dataset_loader.cpp

Line 25 in 13ed38c

SetHeader(filename);

2. Apart from ignore_column, parameters two_round, header, label_column, weight_column, group_column also have no effect when loading from binary. I changed the doc. Need to confirm whether we should add warning for these parameters.

[StrikerRUS]

@hzy46 Thanks for your contribution! Parameters.rst shouldn't be touched directly. Instead, make your changes in the config.h file and then run python helpers/parameter_generator.py to update Parameters.rst.
Refer to https://github.com/microsoft/LightGBM/blame/master/docs/Parameters.rst#L1.

[hzy46]

@hzy46 Thanks for your contribution! Parameters.rst shouldn't be touched directly. Instead, make your changes in the config.h file and then run python helpers/parameter_generator.py to update Parameters.rst. Refer to https://github.com/microsoft/LightGBM/blame/master/docs/Parameters.rst#L1.

OK, got it.

[shiyu1994]

The testing cases are failing because SetHeader does not only handle cases where input are from files. It also reads categorical feature indices from the config parameters (see the part outside the if (filename != nullptr) { ... }).

Skipping SetHeader directly here will cause errors when we load data from numpy or pandas arrays (where filename == nullptr) and use categorical features.

So I think we should move the the check filename != nullptr && CheckCanLoadFromBin(filename) == "" into SetHeader. That is, we change if (filename != nullptr) { ... } into if (filename != nullptr && CheckCanLoadFromBin(filename) == "") { ... }

[hzy46]

@shiyu1994 @StrikerRUS Thanks for your comment. I have updated the code. Please check.

[StrikerRUS]

Thank you! Just one question below:

[StrikerRUS]

Just curious, why there are no similar checks for config_.two_round and config_.header?

[hzy46]

I think it should be obvious that two_round and header is not useful when loading from a binary file. What about your opinion?

[StrikerRUS]

I'm going to agree with you about that header is irrelevant param when loading from binary, but I think two_round isn't so obvious. I'd better list all parameters here for the consistency despite whether it is obvious or not that they are not useful.

[shiyu1994]

Can we merge this PR now? If there's any other parameters need a similar explanation, maybe we can leave it in another PR. :)

[StrikerRUS]

OK.
Only header and two_round have notes in their description about that they

works only in case of loading data directly from text file

but providing them for binary file doesn't trigger this warning.

[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.