Installation¶
The supported public setup documented here targets the current main source tree. The package metadata still reports version 0.1.0, but main contains functionality that is not present in the older v0.1.0 tag or its archived packages. The hemispec-toolkit project is not currently public on PyPI.
Recommended source install¶
Git LFS is required to retrieve the model files tracked under assets/models/:
git lfs install
git clone https://github.com/mqqq333/HemiSpec.git
cd HemiSpec
git lfs pull
python -m pip install -e ".[gui,model,classifier]"
git rev-parse HEAD
Record the output of git rev-parse HEAD with each analysis. A branch name and the package version alone do not identify the source revision reproducibly.
PyTorch must be installed in the same Python or conda environment used to launch HemiSpec. Configure the appropriate CPU or CUDA PyTorch build before a model run.
The checked-out DGN and classifier assets are used directly from assets/models/. For classifier validation, use these local classifier assets or provide an explicit approved local classifier directory; do not rely on a cache pre-download as part of the recommended setup.
Archived v0.1.0 release¶
The GitHub v0.1.0 release archives the original wheel, source distribution, and Windows artifacts. It is a historical release, not a package of current main features. In particular, the v0.1.0 tag does not contain the current synthetic quickstart or model-cache downloader modules.
After downloading the archived wheel, its base CLI can be inspected with:
Do not use the archived wheel as the installation path for current quickstart, model discovery, or model download documentation. See Release artifacts for the exact archive contents.
Development install¶
From a current source checkout:
python -m pip install -e ".[dev,gui]"
python -m pytest
python -m ruff check src tests
python -m mkdocs build --strict
The distribution name is hemispec-toolkit; the import path and CLI command are hemispec. Documentation should not present a PyPI install command until the project is actually public there.
Neuroimaging prerequisites¶
The model-enabled workflow starts from preprocessed GM maps, not raw T1 images. The repository scripts depend on FSL tools including BET, FAST, FLIRT, and fslmaths to produce MNI152 1.5 mm *_GM_masked.nii.gz inputs.
Read Input and preprocessing before processing real data.
GUI and model runtime¶
Launch hemispec-gui from the source environment containing PyTorch. HemiSpec discovers assets from explicit paths, environment variables, the Git-LFS checkout under assets/models/, or the per-user cache. Wheels and lightweight Windows artifacts do not embed PyTorch or the 300 MB+ DGN checkpoints. See Data and models for the current asset boundary.