EEG Classification using Deep and Transfer Learning

Requirements

Dependencies can be found in:

• requirements.txt
• environment.yml (for conda virt env)

Data:

Download EEG MI datasets from:

• BCI Competition IV 2a: http://www.bbci.de/competition/iv/#dataset2a
• High-Gamma: https://web.gin.g-node.org/robintibor/high-gamma-dataset

Create dirs:

In the root of the project:

• Make a directory for the (processed) data: "/data"
• Make a directory for the results: "/results"

Data preprocessing:

To preprocess the data got to "/data_loader/process_data/.." For both datasets there is separate preprocess script.

Config the experiment:

In "config.yaml" (in the project root) you can configure the experiment.

• experiment->dataset: [hgd, bciciv2a]
• model->name: [eegnet, deep, shallow]
• experiment->type: [no_tl, loo_tl, ccsa_da] *no_tl: no transfer learning, just dataset+CNN model *loo_tl: leave-one-out transfer leaning *ccsa_da: feature space alignment, siamese network with contrastive loss, ccsa: classification and contrastive semantic alignment loss
• server->full_cv: True=trains model for all subjects 4-fold (sequentially), False=trains only model for subject_id and i_valid_fold given.

Run experiment:

Run the file "/experiment/run.py" to start the experiment. Results get put into "/results/" dir.

🎩