[python-package] support GPU training with the conda-forge package

[BinchaoPeng]

Summary

I use conda install -c conda-forge lightgbm to install lightgbm, and I found it may not support gpu device.

Motivation

Installing gpu-version lightgbm by conda command.

Description

when conda install -c conda-forge lightgbm, then

LightGBMError: GPU Tree Learner was not enabled in this build. Please recompile with CMake option -DUSE_GPU=1

Therefore, is it possible to get gpu lightgbm with conda install?

References

[github_使用conda安装](https://github.com/Microsoft/LightGBM/tree/master/python-package#install-from-conda-forge-channel)

[BinchaoPeng]

I think the simplest way to use lightgbm that is just use conda install lightgbm and then I can use lightgbm as like pytorch.However, it semms that lightgbm does not has the feature.

[jameslamb]

@BinchaoPeng thanks for using LightGBM!

The conda-forge package for lightgbm is not maintained by this project's maintainers (https://github.com/conda-forge/lightgbm-feedstock) and does not currently support GPU training.

We can take it as a feature request to try to add that support there at some point, and it might get easier in the future with the work done in #3403 and #5252 .

It looks like there is someone trying to add integrated GPU support to the lightgbmpackage on conda-forge (conda-forge/lightgbm-feedstock#40), but that PR so far hasn't received feedback from that feedstock's maintainers.

[BinchaoPeng]

Thank you very much!

[jameslamb]

Sure! I'm going to close this for now, as we keep issues for feature requests closed (and track them in #2302) in this repo until they're being actively worked on.

Thanks again for your interest in LightGBM!

[jameslamb]

Sorry, this was locked accidentally. Just unlocked it. We'd still love help with this feature!

[StrikerRUS]

Is this solved according to the following?

• GPU-enabled package builds conda-forge/lightgbm-feedstock#50
• OpenCL and GPU builds conda-forge/lightgbm-feedstock#55
• https://anaconda.org/conda-forge/lightgbm/files?version=4.4.0
  

[jameslamb]

I think so! I'd like to test it on a system with CUDA when I have time and update LightGBM's documentation.

I think the CUDA-based version will automatically be chosen on systems with CUDA available, thanks to the __cuda metapackage from conda, and otherwise the OpenCL-based version will be installed.

[StrikerRUS]

I'd like to test it on a system with CUDA

Just did it:

~$ conda install lightgbm
Retrieving notices: ...working... done
Channels:
 - conda-forge
Platform: linux-64
Collecting package metadata (repodata.json): done
Solving environment: done


==> WARNING: A newer version of conda exists. <==
    current version: 24.3.0
    latest version: 24.5.0

Please update conda by running

    $ conda update -n base -c conda-forge conda



## Package Plan ##

  environment location: /opt/miniforge3/envs/myenv

  added / updated specs:
    - lightgbm


The following packages will be downloaded:

    package                    |            build
    ---------------------------|-----------------
    cudatoolkit-11.8.0         |      h4ba93d1_13       682.5 MB  conda-forge
    libgfortran-ng-14.1.0      |       h69a702a_0          49 KB  conda-forge
    libgfortran5-14.1.0        |       hc5f4f2c_0         1.4 MB  conda-forge
    libopenblas-0.3.27         |pthreads_hac2b453_1         5.3 MB  conda-forge
    libstdcxx-ng-14.1.0        |       hc0a3c3a_0         3.7 MB  conda-forge
    lightgbm-4.4.0             |cuda_py3.10hc57aca7_3         6.6 MB  conda-forge
    numpy-2.0.0                |  py310h515e003_0         7.4 MB  conda-forge
    python_abi-3.10            |          4_cp310           6 KB  conda-forge
    scipy-1.14.0               |  py310h93e2701_1        16.0 MB  conda-forge
    ------------------------------------------------------------
                                           Total:       722.9 MB

The following NEW packages will be INSTALLED:

  cudatoolkit        conda-forge/linux-64::cudatoolkit-11.8.0-h4ba93d1_13
  libblas            conda-forge/linux-64::libblas-3.9.0-22_linux64_openblas
  libcblas           conda-forge/linux-64::libcblas-3.9.0-22_linux64_openblas
  libgfortran-ng     conda-forge/linux-64::libgfortran-ng-14.1.0-h69a702a_0
  libgfortran5       conda-forge/linux-64::libgfortran5-14.1.0-hc5f4f2c_0
  liblapack          conda-forge/linux-64::liblapack-3.9.0-22_linux64_openblas
  libopenblas        conda-forge/linux-64::libopenblas-0.3.27-pthreads_hac2b453_1
  libstdcxx-ng       conda-forge/linux-64::libstdcxx-ng-14.1.0-hc0a3c3a_0
  lightgbm           conda-forge/linux-64::lightgbm-4.4.0-cuda_py3.10hc57aca7_3
  numpy              conda-forge/linux-64::numpy-2.0.0-py310h515e003_0
  python_abi         conda-forge/linux-64::python_abi-3.10-4_cp310
  scipy              conda-forge/linux-64::scipy-1.14.0-py310h93e2701_1


Proceed ([y]/n)? y


Downloading and Extracting Packages:

Preparing transaction: done
Verifying transaction: done
Executing transaction: \ By downloading and using the CUDA Toolkit conda packages, you accept the terms and conditions of the CUDA End User License Agreement (EULA): https://docs.nvidia.com/cuda/eula/index.html
                                                                                                                                       done

And after installation CUDA works 🎉 :

~$ python
Python 3.10.14 | packaged by conda-forge | (main, Mar 20 2024, 12:45:18) [GCC 12.3.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import lightgbm as lgb
>>> import numpy as np
>>> rng = np.random.default_rng()
>>> data = rng.uniform(size=(500, 10))  # 500 entities, each contains 10 features
>>> label = rng.integers(low=0, high=2, size=(500, ))  # binary target
>>> lgb.LGBMClassifier(device='cuda').fit(data, label)
[LightGBM] [Warning] Using sparse features with CUDA is currently not supported.
[LightGBM] [Info] Number of positive: 227, number of negative: 273
[LightGBM] [Info] Total Bins 1670
[LightGBM] [Info] Number of data points in the train set: 500, number of used features: 10
[LightGBM] [Info] [binary:BoostFromScore]: pavg=0.454000 -> initscore=-0.184522
[LightGBM] [Info] Start training from score -0.184522
[LightGBM] [Warning] No further splits with positive gain, training stopped with 20 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 19 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 18 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 19 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 19 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 20 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 20 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 19 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 18 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 19 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 18 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 18 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 20 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 21 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 20 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 19 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 20 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 18 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 19 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 18 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 19 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 20 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 20 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 20 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 19 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 18 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 19 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 18 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 20 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 19 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 19 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 20 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 18 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 18 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 18 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 18 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 20 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 19 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 20 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 18 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 18 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 19 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 20 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 20 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 18 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 19 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 20 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 21 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 19 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 18 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 18 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 20 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 21 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 20 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 19 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 19 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 19 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 19 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 16 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 18 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 18 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 21 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 21 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 19 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 19 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 18 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 20 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 18 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 19 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 19 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 19 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 20 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 18 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 20 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 17 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 19 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 20 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 20 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 20 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 20 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 21 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 19 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 19 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 20 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 20 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 19 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 21 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 19 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 20 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 19 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 21 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 19 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 18 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 21 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 21 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 19 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 21 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 20 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 20 leaves.
[LightGBM] [Warning] No further splits with positive gain, training stopped with 20 leaves.
LGBMClassifier(device='cuda')

Note the following line from logs: [LightGBM] [Warning] Using sparse features with CUDA is currently not supported.

[jameslamb]

Hey great! Ok yes I think we can consider this solved then.

For those finding this from search engines and looking for something to copy-paste:

conda install -c conda-forge 'lightgbm>=4.4.0'

I will put up some docs changes.

[BinchaoPeng]

So cool, respect!🥴