Setup

Environment

1. Initial env setup

conda env create -f environment.yaml -y

conda activate graph

2. PyTorch setup: need to install the correct packages with the correct cpu/cuda wheel. In our case:

pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu121

pip install torcheeg

Dataset

1. Download the HMS dataset from Kaggle (note: you will have to join the challenge and accept T&C in order to dowload)

cd data/raw

kaggle competitions download -c hms-harmful-brain-activity-classification

2. Run the preprocessing notebook

jupyter execute notebooks/eda.ipynb

3. Create the Graph Dataset

python src/data/make_graph_dataset.py

Training + Experiments

1. AlphaHMS:

python src/train.py --train-config configs/train.yaml

2. Baseline EEG

python src/train.py --train-config configs/train.yaml

3. Baseline MLP

python src/train_mlp.py --config configs/training_mlp.yaml

Training scripts use WandB as logger; you may be asked to log into your account beforehand. Evaluation of the model is performed at the end of the training stage.