Installation¶

Requirements¶

Python 3.11 or later
PyTorch 2.0 or later (the only runtime dependency)

From GitHub¶

pip install "yetanotherspdnet @ git+https://github.com/Yet-Another-Research-Organisation/yetanotherspdnet.git"

A branch or a tag can be pinned with ...yetanotherspdnet.git@<ref>.

From source, for development¶

git clone https://github.com/Yet-Another-Research-Organisation/yetanotherspdnet.git
cd yetanotherspdnet
pip install -e ".[all]"   # test + dev + docs extras

The extras can also be installed separately:

Extra	Contents
`test`	pytest, pytest-cov, SciPy (reference implementations used by the tests)
`dev`	ruff (pinned, same version as CI), mypy, pre-commit
`docs`	Sphinx, MyST, Furo theme

Check the installation¶

import torch
import yetanotherspdnet
from yetanotherspdnet import SPDnet
from yetanotherspdnet.random.spd import random_SPD

print(yetanotherspdnet.__version__)
X = random_SPD(n_features=8, n_matrices=4, dtype=torch.float64)
print(SPDnet(input_dim=8, hidden_layers_size=[4], output_dim=2)(X).shape)
# torch.Size([4, 2])

Run the tests¶

pytest                                   # full suite, coverage enforced (>= 40%)
pytest tests/nn/test_base.py::TestBiMap --no-cov  # a single class

PyTorch builds¶

If the default PyTorch wheel does not suit your machine, install PyTorch first from the official selector, then install this package.