Require PyTorch dependency for reticulate when installing a package

[johnG]

I want to include PyTorch as a requirement for an R library using the reticulate -package.

(Another solution would be if it is installed (i.e., reticulate::py_install("PyTorch") ) when the user is using the library/function the first time, but not every time.)

I tried to add PyTorch as a dependence through the "header"/"description" of the function:

#' @importFrom reticulate PyTorch

But when I run devtools::document() I get this error:

Loading rpackageX Error in asChar(ivars) : empty name in directive 'importFrom' in 'NAMESPACE' file

Thanks in advance.

[jdblischak]

@johnG

I tried to add PyTorch as a dependence through the "header"/"description" of the function:

The roxygen2 tag @importFrom is for declaring R package dependencies. Since PyTorch is a Python package, that won't work.

Here are some resources to help you decide how to handle the PyTorch dependency:

• The reticulate package has a vignette titled Using reticulate in an R Package that describes some best practices.

• The reticulate package has a vignette titled Managing an R Package’s Python Dependencies. It describes how you can list Python packages in a special field called Config/reticulate, which directs reticulate to automatically install these Python packages.

• The tensorflow R package uses reticulate to wrap the tensorflow Python package. It does the following:

  • Provides the function install_tensorflow()
  • Lists Tensorflow in the SystemRequirements field
  • Lists tensorflow in the Config/reticulate field

• The keras R package uses reticulate to wrap the keras Python package. It does the following:

  • Provides the function install_keras()
  • Lists Keras in the SystemRequirements field