This repository contains an example of the code to load the SeizeIT2 dataset and to train the model included in the dataset paper.
Script with an example for loading files from the dataset. The classes classes.data and classes.annotation are used to create a data object, containing the signal data and extra information, and an annotation object, containing all information regarding the seizure events of the recording.
Script to train and evaluate the ChronoNet model with all parameters as in the paper. This is a suggestion of a framework that uses the data loaders and a Keras implementation of the training and evaluation routines. The data generators are likely to take a long time to run (arround 3 hours), hence the option to save the training and validation generators and load them in future runs.
The python packages (and corresponding versions) used in the development of the scripts in this repository are gathered in 'environment.yml'. To easily create a conda environment with the same package versions to run the code, follow the instructions below:
conda config --add channels conda-forge
conda config --set channel_priority strict
conda env create -n ENV_NAME -f environment.yml