Guolin Ke

It is mostly caused by the mis-used of this api. please refer to this test: https://github.com/microsoft/LightGBM/blob/252828fd86627d7405021c3377534d6a8239dd69/tests/c_api_test/test_.py#L124-L154

are there any tests for the multi-GPU training?

Your contribution is very welcomed!

yeah, the training is sensitive to the RAM speed. The core algorithm (histogram accumulation) in LightGBM is quite simple and just contains additions. The bottleneck is indeed the memory speed.

@dthiagarajan It seems there are some problems when constructing Dataset distributedly. Can you provide the number of columns (features) of the used dataset?

> I also found that the script works when I change the tree_learner to voting from data and when I add num_threads explicitly to be the core count on each...

sorry for missing this PR. I think this PR indeed fixes the bug. cc @shiyu1994 to confirm it.

It seems we need to fix some tests

@shiyu1994 can you help to review this PR?

@junpeng0715 can you also add a CI test, to build with int64_t type and run some tests?