HPFRACC development environment (summary)๏
This file mirrors the canonical setup in config/environment.yml and CONTRIBUTING.md. Prefer those sources if anything conflicts.
Overview๏
Conda env name:
hpfracc(legacy nameshpfracc-env/fracnnmay still exist on older machines)Python: 3.11 (recommended; CI smoke / full pytest / RTD use 3.11; main test workflow also runs 3.10 and 3.12)
Install: editable checkout with dev extras
Quick start๏
conda env create -f config/environment.yml # or: conda env update -f config/environment.yml
conda activate hpfracc
pip install -e ".[dev]"
Optional shell helpers (from repo root):
source tools/activate_env.sh
Core stack (from config/environment.yml)๏
Versions drift over time; the YAML is authoritative. In general the env includes NumPy, SciPy, Matplotlib, Pandas, SymPy, JAX/jaxlib, Numba, PyTorch (via pip), Optax, pytest stack, Sphinx, Black, Flake8, Mypy, and pre-commit.
Testing๏
conda activate hpfracc
python -c "import hpfracc; print(hpfracc.__version__)"
python -m pytest tests/ -c pytest.ini --tb=short
Managing the environment๏
conda env list
# Recreate from repo spec (do not overwrite environment.yml with conda export unless you intend to pin everything):
conda env remove -n hpfracc # only if you need a clean rebuild
conda env create -f config/environment.yml
Troubleshooting๏
CUDA / GPU: match driver and optional
jax[cuda12]/ PyTorch CUDA builds to your hardware (see README).Import errors: confirm
conda activate hpfraccand thatpip install -e ".[dev]"was run from the repository root.
Performance๏
GPU memory and throughput depend on your hardware; use nvidia-smi (NVIDIA) or smaller batch sizes if you hit OOM during ML workloads.