Interpreting neural decoding models using grouped model reliance
This repository provides code and preprocessed data for the analyses presented in Valentin, Harkotte and Popov (2020).
- EEG preprocessing and matlab-based analyses are provided in
- The implementation of grouped model reliance is provided in
- Feature groups, provided as indices of columns of the data matrices, can be found in
- The script
compute_mr.pycomputes model reliance scores for both frequency and regions of interest and saves them to the
results/directory. For faster computation, consider lowering the number of trees for the Random Forest algorithm and/or the number of cross validation folds and random permutations. Running
python compute_mr.pysaves the results in
- The main figures are generated in
- Analyses for supplementary figures are provided in separate notebooks as
Preprocessed data matrices for each participant (with anonymised IDs) can be found under
data/. Rows correspond to trials, and columns to power per frequency (in 1 Hz bins) per electrode location.