The program crashed when using the log1p dataset. about thundergbm HOT 5 OPEN

Would you please provide more information (e.g., OS, CUDA version, GPU, which interface) to help us reproduce the problem? It would be good to upload a minimum example to help us reproduce the problem efficiently.

from thundergbm.

Would you please provide more information (e.g., OS, CUDA version, GPU, which interface) to help us reproduce the problem? It would be good to upload a minimum example to help us reproduce the problem efficiently.

Ubuntu 16.04, CUDA10.1.168, GTX970, I just download and built the cude follow the instruction in readme. Run with the quick start command with changing the dataset location.

from thundergbm.

log1p is a large dataset and the training process requires more memory than your GTX970 has. A suggestion is to sample the log1p data set, and obtain a subset which your GPU can process.

from thundergbm.

log1p is a large dataset and the training process requires more memory than your GTX970 has. A suggestion is to sample the log1p data set, and obtain a subset which your GPU can process.

thx, I will try it. But isn't it supposed to be out of memory error? Just like when I try the higgs dataset.

from thundergbm.

Thanks. We will improve the output in the future patch.

from thundergbm.