Installation

Requires Python 3.10–3.13. The base installation depends only on NumPy:

python -m pip install renewable-huber
renewable-huber --version

Optional Rust CPU core

The native CPU core is distributed separately from the pure-Python base package. Once renewable-huber-native-cpu is installed or built locally, you can select it explicitly, or let CPU auto select it when its conservative measurement passes.

Install the matching native wheel directly; it depends on exactly the same version of the base package:

python -m pip install renewable-huber-native-cpu==0.7.1

The 0.7.1 release wheels cover CPython 3.10–3.13, Windows x86-64, Linux x86-64/aarch64 and macOS x86-64/Apple Silicon. Regular users do not need to install Rust or compile the extension locally.

How to select the engine, and how n_jobs sizes its thread pool, is covered in Backends.

Optional native CUDA engine

The CUDA 12 plugin wheel can be installed directly. 0.7.1 provides Windows x86-64 and Linux x86-64 (manylinux_2_28) wheels for CPython 3.10–3.13:

python -m pip install renewable-huber-native-cuda==0.7.1

The wheel already contains the native extension compiled for the supported GPU architectures, so Rust, CMake, Visual Studio and a local nvcc are not needed. The wheel itself does not bundle the NVIDIA libraries: the CUDA 12 runtime closure (cudart, cuBLAS/cuBLASLt, cuSOLVER, cuSPARSE and nvJitLink) is installed automatically as dependencies from NVIDIA's official nvidia-*-cu12 wheels, and that set is loaded first at import time. Only when the set is incomplete does it fall back to the system toolkit (CUDA_PATH on Windows, the default loader path on Linux), so CUDA_PATH does not need to be set. A compatible NVIDIA driver is still required.

Using the engine is covered in Backends.

Extras

Install the extra that matches your use case:

Use case	Install command
pandas input	python -m pip install "renewable-huber[pandas]"
scikit-learn adapter	python -m pip install "renewable-huber[sklearn]"
CuPy / CUDA 12	python -m pip install "renewable-huber[gpu-cupy]"
PyTorch	python -m pip install "renewable-huber[gpu-torch]"
TensorFlow	python -m pip install "renewable-huber[gpu-tensorflow]"

The GPU extras install only the corresponding framework; they do not install the NVIDIA driver or the CUDA runtime for you. First check compatibility between the framework, operating system, Python and GPU driver; see the support matrix for detailed limits.

From source

For development from source:

python -m pip install -e ".[dev]"

Building the native extensions locally, and the WSL2/Linux toolchain profiles, are described in Contributing.