Dialog box pops up when using Debug build of LightGBM

[mjmckp]

When I attempt to use a debug build of LightGBM, when I create the booster object with default parameters, I get the following dialog box pop up:

This doesn't happen with a release build. I think this is being generated by OpenMP, I've seen mention of it elsewhere in projects that use OpenMP.

Steps to reproduce:

• Clone LightGBM.Net from https://github.com/rca22/LightGBM.Net
• Switch to the update_jun_20 branch (No longer necessary, use master branch)
• Put a Debug build of LightGBM (built from master) in the LightGBM.Net\native\Debug folder
• Run the test BoosterTest.Create in LightGBMNet.Train.Test

[jameslamb]

Thanks for the report and for using LightGBM!

We do not maintain LightGBM.NET, could you please raise this issue in that project's issues page?

If you are able to reproduce this pop up using only LightGBM code here, please open a new issue and we'll be happy to help.

[mjmckp]

Sorry, I think you have misunderstood, the issue has nothing to do with LightGBM.Net, I maintain that project, and I am just using it as a way to explain how to replicate the issue. The error is being generated by the native lightgbm dll.

[jameslamb]

oh I definitely did misunderstand. Can you please provide a reproducible example that doesn't rely on LightGBM.Net? That would help us to help you

[guolinke]

@mjmckp The `Debug' is the 'debug' mode of lightgbm.dll or LightGBM.net?

[mjmckp]

@guolinke The debug build of lightgbm.dll

[shiyu1994]

@mjmckp This is due to a bug in undefined "num_threads" of the TrainingShareStates. The num_threads of a TrainingShareStates is set in this function. However, when the bin is nullptr, num_threads of TrainingShareStates is not set.

LightGBM/include/LightGBM/dataset.h

Lines 296 to 308 in f5f27ca

	void SetMultiValBin(MultiValBin* bin) {
	if (bin == nullptr) {
	return;
	}
	multi_val_bin.reset(bin);
	num_threads = OMP_NUM_THREADS();
	num_bin_aligned =
	(bin->num_bin() + kAlignedSize - 1) / kAlignedSize * kAlignedSize;
	size_t new_size = static_cast<size_t>(num_bin_aligned) * 2 * num_threads;
	if (new_size > hist_buf.size()) {
	hist_buf.resize(static_cast<size_t>(num_bin_aligned) * 2 * num_threads);
	}
	}

But later on, this num_threads will be passed to num_threads argument of omp here

LightGBM/src/io/dataset.cpp

Line 1336 in f5f27ca

#pragma omp parallel for schedule(static) num_threads(share_state->num_threads)

which causes the error.

Inserting num_threads = OMP_NUM_THREADS(); between line 297 and 298 of dataset.h resolve this issue.

We will resolve this with a pull request together with the problem in #3152.

[guolinke]

@shiyu1994 I think you can create a PR to fix this first.

[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.