Using chb-mit dataset from https://physionet.org/content/chbmit/1.0.0/ Warning! In the file RECORDS-WITH-SEIZURES in line 35 (chb07/chb07_18.edf) should be changed into chb07/chb07_19.edf How to use this repo:
- Download chb-mit dataset using link from above into repo folder, pull the directory containing data as a main directory and rename it to raw_dataset
- Create python environment and install requirements.txt.
- To run preprocessing please execute the scrip preprocessing/run_preprocessing.py
Note that preprocessing parameters are hardcoded in utils/utils.py file
to change them they need to be configured manually. - To run training please execute the script train.py. By default wandb
is enabled, to run it as is one needs to create wandb_api_key.txt file in
src folder with wandb API key. - If one wants to replicate the wandb sweep for architecture search,
please refer to the instructions on https://wandb.ai/. Sweep file is in
sweep_config.yaml. To run the sweep please modify the parameters, such as
entity or model configuration. - To run explainability notebook, please follow the instructions in the notebook
named explainability.ipynb.
Publications based on this work:
Mazurek, S., Blanco, R., Falcó-Roget, J., Argasiński, J.K., Crimi, A. (2023).
Impact of the Pre-processing and Balancing of EEG Data on the Performance of Graph Neural Network for Epileptic Seizure Classification.
In: Rutkowski, L., Scherer, R., Korytkowski, M., Pedrycz, W., Tadeusiewicz, R., Zurada, J.M. (eds)
Artificial Intelligence and Soft Computing. ICAISC 2023. Lecture Notes in Computer Science(), vol 14126. Springer, Cham.
https://doi.org/10.1007/978-3-031-42508-0_24