Installation

Requirements

• Python >= 3.9
• NumPy >= 1.20
• Pandas >= 1.3
• SciPy >= 1.7
• h5py >= 3.0
• anndata >= 0.8
• scanpy >= 1.9

Optional: CuPy >= 10.0 (GPU acceleration)

Virtual Environment

Create a virtual environment before installing to avoid dependency conflicts:

python -m venv secactpy-env
source secactpy-env/bin/activate   # Linux/macOS
# secactpy-env\Scripts\activate    # Windows

From PyPI (Recommended)

# CPU Only
pip install secactpy

# With GPU Support (CUDA 11.x)
pip install "secactpy[gpu]"

# With GPU Support (CUDA 12.x)
pip install secactpy
pip install cupy-cuda12x

From GitHub

# CPU Only
pip install git+https://github.com/data2intelligence/SecActpy.git

# With GPU Support (CUDA 11.x)
pip install "secactpy[gpu] @ git+https://github.com/data2intelligence/SecActpy.git"

# With GPU Support (CUDA 12.x)
pip install git+https://github.com/data2intelligence/SecActpy.git
pip install cupy-cuda12x

Development Installation

git clone https://github.com/data2intelligence/SecActpy.git
cd SecActpy
pip install -e ".[dev]"

Documentation Tools

To build the documentation site locally:

pip install -e ".[docs]"
mkdocs serve

Native R install (Linux / macOS / Windows)

SecActPy is a pure-Python package and runs on its own. The optional R-side stack (SecAct + accelerators) is needed only when you want to run the original R implementation alongside SecActPy for cross-validation or benchmarking.

The R packages are:

Package	Role	Required	Platforms
SecAct (data2intelligence/SecAct)	R-native inference (no compiled deps)	Yes	Linux / macOS / Windows
RidgeFast (data2intelligence/RidgeFast)	Optional CPU accelerator (R + C, GSL)	No	Linux / macOS / Windows
RidgeCuda (data2intelligence/RidgeCuda)	Optional GPU accelerator (R + CUDA)	No	Linux + NVIDIA only

The legacy beibeiru/RidgeR package is archived and is not used by SecActPy anymore.

Linux (Debian / Ubuntu)

# 1. R + system libraries (GSL for RidgeFast, OpenBLAS for BLAS, TBB for OpenMP)
sudo apt update
sudo apt install -y r-base r-base-dev libgsl-dev libopenblas-dev libtbb-dev

# 2. R packages
Rscript -e 'install.packages(c("remotes", "BiocManager"))'
Rscript -e 'remotes::install_github(c(
  "data2intelligence/SecAct",
  "data2intelligence/RidgeFast"   # optional CPU accelerator
))'

# 3. (Optional, NVIDIA GPU only) RidgeCuda
# Requires CUDA Toolkit >= 11.0 with cuBLAS / cuSOLVER / cuSPARSE / cuRAND.
Rscript -e 'remotes::install_github("data2intelligence/RidgeCuda")'

macOS

# 1. Toolchain via Homebrew
brew install r gsl libomp openblas

# 2. R packages
Rscript -e 'install.packages(c("remotes", "BiocManager"))'
Rscript -e 'remotes::install_github(c(
  "data2intelligence/SecAct",
  "data2intelligence/RidgeFast"
))'

Note

RidgeCuda is not supported on macOS — Apple does not ship NVIDIA drivers. Use SecAct (R-native) or SecActPy with backend="cupy" from a Linux machine for GPU inference.

Windows

1. Install R from https://cloud.r-project.org and Rtools43+ from https://cran.r-project.org/bin/windows/Rtools/ (Rtools provides the compiler toolchain plus a bundled pacman for system libraries).
2. From the Rtools mingw64 / ucrt64 shell, install GSL:

pacman -S mingw-w64-ucrt-x86_64-gsl

1. From a regular R session:

install.packages(c("remotes", "BiocManager"))
remotes::install_github(c(
  "data2intelligence/SecAct",
  "data2intelligence/RidgeFast"
))

Note

RidgeCuda is not supported on Windows. The Linux+NVIDIA combination is the only supported configuration for GPU R-side inference.

Verify the install

library(SecAct)
library(RidgeFast)        # if installed
# library(RidgeCuda)      # GPU only

cat("R version:", R.version.string, "\n")
cat("SecAct:",    as.character(packageVersion("SecAct")),    "\n")
cat("RidgeFast:", as.character(packageVersion("RidgeFast")), "\n")

If you skip RidgeFast, SecAct falls back to its pure-R ridge path. That path is correct but slower; the CPU accelerator is recommended for any non-trivial workload.