search

rocker-org / ml

experimental machine learning container
GNU General Public License v2.0
50 stars 13 forks source link

Build xgboost for R with multi-GPU support #11

Closed noamross closed 5 years ago

noamross commented 5 years ago

One little tweak required to let the xgboost R package run on multiple GPUs. Adding a commented-out multi-GPU that runs with this change.

Everything in the test script runs fine on my hardware (2X GTX 1080, 2X Titan XP) .

I also attempted to fit a multi-GPU Keras model (https://keras.rstudio.com/reference/multi_gpu_model.html). This had issues because of an apparent bug in 1.12 compatibility (https://github.com/rstudio/tensorflow/issues/272). Reverting to 1.10 detected all my GPUs but there was some memory issue. The R package didn't detect tensorflow when I trieed installing 1.13rc2 .

cboettig commented 5 years ago

Thanks! Speaking of the xgboost install, can you get it to compile from a fixed release, rather than the latest build? (I tried just unpacking https://github.com/dmlc/xgboost/archive/v0.81.tar.gz but it does not seem to build for me with this recipe...)