Installation

From PyPI, from the first release (0.3.0) on:

pip install dietnetwork
dietnet info

From a clone, with the locked versions:

git clone https://github.com/MattScicluna/DIETNETWORK && cd DIETNETWORK
uv sync --frozen          # or: pip install -e .
uv run dietnet info

Or straight from GitHub, without a clone:

pip install git+https://github.com/MattScicluna/DIETNETWORK

The wheel carries HussinLab's Dietnet_train_and_evaluate/ inside the package (as dietnet/_upstream/), byte for byte, so an installed dietnet runs exactly the published training and model code.

Tested on Python 3.10–3.13, on Linux and macOS. A GPU is optional: CPU is enough for prediction, and training uses CUDA when torch finds it.

Checking it works

dietnet info

prints the package and torch versions, whether CUDA is available, and the upstream commit it runs. It also hashes every upstream file against src/dietnet/upstream_manifest.json, and exits with status 1, naming the files, if any is modified, missing or added:

dietnet 0.3.0
upstream HussinLab/DIETNETWORK 5481556e472140c6d3a67363c1286f22109fb1ce at ...
upstream files: identical to the pinned commit
torch ..., cuda available: False

The test suite runs the whole pipeline on small fixtures in a few minutes:

uv sync --frozen --extra dev
uv run pytest -m "not slow and not network"

The parity tests skip unless a full upstream checkout is given; see parity with upstream.

Extras

extra	for
alphagenome	dietnet alphagenome: the hosted AlphaGenome client
tracking	comet-ml
docs	this site
dev	tests, formatters, pre-commit

uv sync --frozen --extra alphagenome      # in a clone, locked versions
pip install -e '.[alphagenome]'           # in a clone, with pip
pip install 'dietnetwork[alphagenome]'                 # from PyPI
pip install 'dietnetwork[alphagenome] @ git+https://github.com/MattScicluna/DIETNETWORK'

Developing on the repository

uv sync --frozen --extra dev && uv run pre-commit install
uv run pytest

black, isort and flake8 run at line length 100. pre-commit also runs a check for files that should not be shared (genotypes, HDF5 datasets, .npy/.npz arrays, .pt weights, cluster paths) and gitleaks. Dietnet_train_and_evaluate/ is upstream's code and is never edited: tests/unit/test_upstream_unchanged.py fails on any change.

To work on this site:

uv sync --frozen --extra docs
uv run mkdocs serve

The tutorial renders without its outputs, because the build does not execute it; the docs workflow does that in a separate step. To see them locally:

uv run jupyter nbconvert --to notebook --inplace --execute docs/tutorial.ipynb

(and do not commit the executed notebook).