Skip to content

Latest commit

 

History

History
44 lines (30 loc) · 1.84 KB

README.md

File metadata and controls

44 lines (30 loc) · 1.84 KB

Environment

Create and activate the conda environment named eeg2text from the environment.yaml file.

conda env create -f environment.yaml
conda activate eeg2text

Data

The datasets are not included in this repository. Please follow the instructions below to download and preprocess the datasets. Following Wang and Ji, 2022, only sentiment and normal reading tasks are used.

The downloading instructions and preprocessing code have been adapted from Open Vocabulary Electroencephalography-To-Text Decoding and Zero-shot Sentiment Classification.

Download the ZuCo Datasets

ZuCo 1.0

Download the files for the following tasks from OSF Storage v1.0:

  • task1-SR
  • task2-NR

NOTE: The files are 43.6 GB, it can take some time to download.

Create the following directories in the repository's root directory:

  • dataset/ZuCo/task1-SR/Matlab_files
  • dataset/ZuCo/task2-NR/Matlab_files

Unzip the downloaded files and move the .mat files to their respective directories.

ZuCo 2.0

Download the file task1-NR from OSF Storage v2.0.

NOTE: The file is 62.2 GB.

Create the directory dataset/ZuCo/task2-NR-2.0/Matlab_files in the repository's root directory, unzip the downloaded file and move the .mat files to the created directory.

Preprocess the Datasets

To preprocess the .mat files run the following command:

bash src/prepare_dataset.sh

NOTE: Please be patient, it can take a bit of time to preprocess the files.

For each task, all .mat files will be converted into a single .pickle file and stored in the following path: dataset/ZuCo/<task_name>/pickle/<task_name>-dataset.pickle.