The main objective of this package is to provide a template to
- use
PyTorch
C++ API from an R package - with
Rcpp
- with test in both R and C++ using
testthat
and the wrapper intestthat
of the C++Catch
library - and which is easy to use (e.g. can use
R CMD INSTALL
,R CMD build
, anddevtools::test()
).
This is instructions for Linux. Download and unzip LibTorch (https://pytorch.org/)
https://download.pytorch.org/libtorch/cu101/libtorch-cxx11-abi-shared-with-deps-1.4.0.zip
Set the following enviroment variable
export TORCH_INSTALL_PREFIX=/your/path/to/libtorch
Install the package
git clone https://github.com/boennecd/Rtorchcpp
R CMD build Rtorchcpp && R CMD INSTALL Rtorchcpp_0.1.0.tar.gz
# or
R CMD INSTALL Rtorchcpp
The first build may fail but gives instruction on what to do.
R CMD build Rtorchcpp && R CMD check --as-cran Rtorchcpp_0.1.0.tar.gz
Or
# assuming that you are in the package directory
devtools::test()
- What about the GPU version?
- Ask the user to create a symlink to
libtorch.so
instead. Otherwise, consider some other alternative (e.g. what is done inrJava
package). - Support more platforms.