Intro
GPU-accelerated algorithms (via the WGPU backend on cubecl which allows you to run the code here on any GPU). You will need to set up your wgpu however, please check the CubeCL book. I do not have access to an Nvidia GPU, but I will aim also to allow for conditional compiling to that backend if Cuda is detected (future problem). The package is designed to support the bixverse package. Additionally, also provides some neural net-based versions of embedding methods for manifoldsR.
Usage
Installation
You will need Rust on your system to have the package working. An installation guide is provided here. There is a bunch of further help written here by the rextendr guys in terms of Rust set up. (bixverse.gpu as bixverse both use rextendr to interface with Rust.) Additionally, in this special case, you will also need the GPU drivers set up properly on your system. Please refer to the CubeCL book in terms of how to ensure WGPU runs on your respective system. On a Unix system you also need OpenBLAS set up for the Ndarray backend; on MacOS it will just use the Accelerate framework.
Linux users
The CPU-accelerated form of the neural net powering parametric UMAP is dependent on OpenBLAS. You will need to have OpenBLAS set up on your system!
Setting up Rust
Steps for installation:
- In the terminal, install Rust
curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh- In R, install rextendr:
install.packages("rextendr")- Finally install bixverse.gpu:
devtools::install_github("https://github.com/GregorLueg/bixverse.gpu")Windows support
If you are using Windows, I am sorry, the tool chain is just very, very painful… I really tried to make this work and maybe there are some hacks in terms of compiling everything to make this work, but I cannot guarantee proper behaviour here due to the dependency with h5 (for reading in h5ad files). If you know how to make this work without several hacks in an easy way, please contact me!
How to use the package.
The package website can be found here. This package is not a stand-alone package, but designed to support the bixverse with GPU-accelerated methods (at the moment with focus on single cell support). If you are however interesting in just using the GPU-accelerated kNN searches, feel free to use the respective rs_ functions for that. Or if you want to train a neural network for UMAP, the package also provides what you need.