This repository contains the implementation of DECIFRA architecture designed for derivation of effective connectivity features from fMRI time series.
conda create -n dcfr python=3.12
conda activate dcfr
conda install pytorch torchvision torchaudio pytorch-cuda=11.3 -c pytorch -c nvidia
pip install -r requirements.txt-
mode:tune- tune mode: run multiple experiments with different hyperparamsexp- experiment mode: run experiments with the best hyperparams found in thetunemode, or with default hyperparamsdefault_HPsis set toTrue
-
model: model for the experiment. Models' config files can be found atsrc/conf/model, and their sourse code is located atsrc/models -
dataset: dataset for the experiments. Datasets' config files can be found atsrc/conf/dataset, and their loading scripts are located atsrc/datasets.
prefix: custom prefix for the project- default prefix is UTC time
- appears in the name of logs directory
HP_path: path to custom hyperparams to loadfollow_splits: path to an experiment with train/validation/test splits that you want to replicate in the new experiments.