An experimental fMRI learning-stage classifier with temporal, causal and physiological-state modules.
git clone https://github.com/Intempus-Inc/learnedSpectrum.git
cd learnedSpectrum
python -m venv .venv
source .venv/bin/activate
python -m pip install -e .The training code assumes CUDA. A CPU training path has not been verified. Dependencies are not locked to a reproduced environment.
learnedSpectrum/: model, data processing, training and visualization modules.notebooks/learnedSpectrum.ipynb: original experiment workflow, with saved outputs cleared.scripts/download.py: downloader for ds000002, ds000011, ds000017 and ds000052.
Datasets must be obtained under their original terms. The downloader uses historical archive URLs; availability and downloaded sizes have not been rechecked. Running python scripts/download.py starts downloads for all four datasets.
This is research code, not a verified reproduction package. The notebook's evaluation call and expected metric fields differ from the current training module. Review those interfaces before running an experiment. There is no command-line training entry point. The downloader only extracts raw files; the notebook expects processed .npy inputs and does not run an end-to-end preprocessing step. Its data paths depend on the working directory.
Learning-stage labels are derived from task and file metadata. The current split operates on files, so it does not establish performance on held-out participants. No new performance or causal claim is made by this cleanup.
Copyright 2024 Teddy Warner. See LICENSE for the existing attribution terms. External datasets retain their own terms.