“LightGBMError: Bug. There should be only one multi-val group.”

[stakas]

Apologies for turning here in a StackOverflow fashion, but I have not found any other kind of LightGBM help, so this is the last resort.

I am loading a series of large binary independently binarized with save_binary LightGBM files and concatenating them into a single file to train GBT with LightGBM, however, train method fails with a cryptic error message:

"LightGBMError: Bug. There should be only one multi-val group."

I haven't found anything anywhere regarding this error and had a peek into C++ code and it seems that there is a fix pending regarding this:

LightGBM/src/io/dataset.cpp

Line 1502 in 0c1c36c

// FIXME: fix the multiple multi-val feature groups, they need to be merged

In the meantime, can somebody advise me how to change my data so that the Dataset contruction would work? Thank you in advance.

My snippet is as follows:

import lightgbm as lgb

my_files = ['p1.bin', 'p2.bin', 'p3.bin']

full_data = lgb.Dataset(my_files[0]).construct()

for f in my_files[1:]:        
    full_data.add_features_from(lgb.Dataset(f).construct())  

feature_name = ['feature_' + str(col) for col in range(full_data.num_feature())]
full_data.feature_name = feature_name

params = {'objective': 'binary', 
         'learning_rate': 0.01,
         'metric': ['binary'],
        }

bst = lgb.train(params = params, 
                train_set = full_data, 
                num_boost_round = 500,
                feature_name = feature_name,
                ) 

[StrikerRUS]

Hello @stakas !

Could you please try this branch?
https://github.com/microsoft/LightGBM/tree/fix-add-features

I hope #2754 will fix this issue.

[stakas]

Ah, I had a hunch that is related to the addition of data sets, just could not get through the woods of C code. Will get some sleep and try fix-add-features branch.

Your help much is appreciated @StrikerRUS !

[StrikerRUS]

Fixed in #2754. Feel free to reopen if the error hasn't gone.

[github-actions]

This issue 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.