What is the best way to stream-line the installation of RAPIDS cuML C++ library for R users?

[yitao-li]

See #103

Observations so far:

• It cannot be something fully automated unless we can somehow ensure the installation of RAPIDS cuML C++ library does not happen automatically on a CRAN machine.
• Building this shared library from source, even if without multi-CPU support, is a time-consuming endeavor -- you can see https://github.com/yitao-li/cuml-installation-notes#build-from-source-without-conda-and-without-multi-gpu-support and try it yourself.
• One can also download pre-built RAPIDS cuML libraries from https://anaconda.org/rapidsai/libcuml/files but users need to ensure the files being downloaded match the CUDA version they currently have. Maybe a more stream-lined solution will need to rely on downloading RAPIDS cuML libraries from https://anaconda.org
• Requiring users to install Conda seems non-ideal. Conda is the only officially supported distribution channel for RAPIDS cuML though (as the Docker image also relies on Conda to download the correct version of RAPIDS cuML)
• Maybe publishing Docker images containing the latest stable version of R, latest cuML-supported version of CUDA, and then the latest RAPIDS cuML C++ library will make things easier for R users who primarily use Docker for repeatable ML tasks (?)