Installation#

This guide provides instructions for setting up Heat as a user or a contributor.

New users#

Package managers (conda, mamba, pixi, etc.)#

The Heat conda-forge build includes all necessary dependencies, including OpenMPI.

conda create --name heat_env
conda activate heat_env
conda install -c conda-forge heat

Standard Python (pip)#

The PyPI build includes all Python dependencies. It requires an MPI implementation to be installed on your system. Heat is typically tested against OpenMPI.

python -m venv heat_env
source heat_env/bin/activate
pip install heat[hdf5,netcdf,zarr]

HPC environments#

Heat is a native HPC library designed to run on large-scale clusters. For production environments, it is recommended to use HPC-native package managers, however pip environments and container runtimes are also an option.

EasyBuild & Spack#

Heat provides configurations for common HPC software stack managers. Refer to the specific repository for the most recent recipes:

  • EasyBuild: Search for the heat software block.

  • Spack: Use spack install py-heat.

  • On JSC systems, Heat is available as a module:

module load GCC OpenMPI heat

Docker#

See our docker README on GitHub.

Verification#

Confirm the installation by running a distributed smoke test to verify tensor splitting across processes:

mpirun -n 2 python -c "import heat as ht; x = ht.arange(10, split=0); print(f'Rank {x.comm.rank} local shape: {x.lshape}')"

This should output something like:

Rank 0 local shape: (5,)
Rank 1 local shape: (5,)

indicating that the tensor was successfully split across the two processes.

New contributors#

Development environment setup#

Contributors should install Heat in editable mode to ensure code changes are reflected immediately. Use one of the following methods to establish a development environment.

Method A: automated dependency management (conda / mamba / pixi)#

For automated handling of MPI and CUDA/ROCm toolkits.

  1. Fork the Heat repository and clone it locally (or, if you have write access, clone the main repository).

  2. Go to the root of the cloned repository and create a new conda environment:

cd heat
conda env create -f scripts/heat_dev.yml
conda activate heat_dev

  1. Install Heat in editable mode:

pip install -e '.[dev]'

Method B: manual dependency management (pip)#

Prerequisite: a functional MPI installation already exists on your system.

  1. Fork the Heat repository and clone it locally (or, if you have write access, clone the main repository).

  2. Create a virtual environment and activate it:

python -m venv heat_dev
source heat_dev/bin/activate

  1. Install Heat in editable mode:

pip install -e '.[dev]'

Repository syncing#

If you cloned a fork, add the main repository as remote for synchronizing in the future using:

git remote add upstream https://github.com/helmholtz-analytics/heat.git

Quality Control#

Install pre-commit hooks to enforce coding standards locally before pushing changes:

pre-commit install

All set!#

For more details on our workflow, see our contributing guidelines. We look forward to your contributions!