Installation¶

specsrbench needs Python 3.10 or newer. It is tested on 3.10, 3.11 and 3.12.

From PyPI¶

pip install specsrbench

From a release¶

Every release also attaches a wheel and an sdist, and each is archived on Zenodo with its own DOI. Use these when you need a specific version pinned to a citable archive:

pip install https://github.com/aryana-haghjoo/specsr-benchmark/releases/download/v0.1.1/specsrbench-0.1.1-py3-none-any.whl

From source¶

git clone https://github.com/aryana-haghjoo/specsr-benchmark
cd specsr-benchmark
pip install -e .

Extras¶

The base install carries what the classical baselines and five of the six figures need — numpy, scipy, matplotlib, pandas, PyWavelets, scikit-image — plus huggingface_hub, so that pip install specsrbench is enough to fetch the tutorial sample and run the whole benchmark on it. Installing a deep-learning stack to run a Wiener filter is a bad trade, so torch is not a dependency.

extra	pulls in	needed by
`tutorial`	jupyterlab	running the tutorial notebook — the data needs no extra
`ml`	torch, `specsr`	`specsrbench build predictions`, and figure 1’s toy CNN
`lsf`	astropy	`specsrbench build lsf`, which reads raw JADES `x1d` products
`dev`	pytest, ruff	the test suite
`docs`	sphinx, furo, myst-parser	this site
`all`	`ml` + `lsf` + `tutorial`

pip install -e '.[all]'

Pinned versions¶

requirements.txt pins the exact environment every cached array and figure was verified in (Python 3.11.13). pyproject.toml resolves looser bounds and is what most people want; use the pins if you need a figure to rebuild identically.

Checking it works¶

specsrbench --version
specsrbench paths          # the directories it will read and write
pytest -m "not slow"       # tests that need data skip cleanly

specsrbench paths is the first thing to run when something cannot be found: it prints the four directories being consulted and whether each exists.