Missing feature_name method

[jsuchal]

We are using this library, but seems like when we load model from file we cannot access the feature names.

In our setup we have a features repository that loads actual data for a model based on feature names, so we have to store the feature names elsewhere. However looking into https://lightgbm.readthedocs.io/en/latest/pythonapi/lightgbm.Booster.html#lightgbm.Booster.feature_name and the saved model file, the feature names are already stored in the model, we are probably only missing a wrapper for native method.

I am not very familiar with FFI so if you have any pointers how to add this I am willing to do a PR on this.

Awesome work on this library wrapper so far!

[ankane]

Hey @jsuchal, thanks for reporting! Added in the commit above. This was very similar to the feature_names method for datasets, but the general approach for adding new methods is to find it in the C API header, attach it in lib/lightgbm/ffi.rb, and then add it to the Ruby class.

[StrikerRUS]

Hey @ankane ! Just want let you know that it is possible to not be limited with 255-length (or any other value) buffer size. It is possible to achieve by repeating the call of C API function in case of returned value in out_buffer_len was larger than buffer_len. Just the same as in model_to_string()
https://github.com/ankane/lightgbm/blob/ba58e984c5e6d80a9b0d9b8b9789abef0c3d2259/lib/lightgbm/booster.rb#L105-L108

Please refer to microsoft/LightGBM#4293.

[ankane]

Thanks @StrikerRUS! Fixed in b5b3e19 (and fixed eval_names in a follow up commit).