In September 2020, the LightGBM team merged an experimental CUDA version.
LightGBM released v3.2 in late March with built-in support for Dask. There is an open issue on LightGBM for Dask GPU support.
Do folks think it might be valuable to include the CUDA version of LightGBM in gbm-bench to track performance of this implementation?
In September 2020, the LightGBM team merged an experimental CUDA version.
LightGBM released v3.2 in late March with built-in support for Dask. There is an open issue on LightGBM for Dask GPU support.
Do folks think it might be valuable to include the CUDA version of LightGBM in gbm-bench to track performance of this implementation?