The best of two worlds: Decoding and source-reconstructing M/EEG oscillatory activity with a unique model

Britta U. Westner & Jean-Rémi King

Repository

This repository contains the simulation and analysis code for the project The best of two worlds: Decoding and source-reconstructing M/EEG oscillatory activity with a unique model.

This paper is soon available as a preprint.

Code

The code in this repository is organized as follows:

Configuration file

• project_settings.py

Simulations

• sensor_decoding.py - runs one realization of sensor space decoding
• source_decoding.py - runs one realization of source space decoding
• decoding_stats.py - runs the whole simulation with 200 realizations, takes snr as input, to be run in parallel for SNRs
• decoding_stats_cv.py - runs a grid search for optimal C parameter, 200 realizations, to be run in parallel for SNRs

Real data analysis

• real_data_faces.py - analysis on the faces data set as shipped with MNE-Python

Generate figures and tables

Figures 2, 4, and 5 as well as Tables 1 and 2 are generated using the Jupyter Notebooks in the subfolder jupyter:

• plotting_statistics_results.ipynb - Fig. 2
• plotting_statistics_results_gridsearch.ipynb - Fig. 4 and 5

Figure 3 is generated using the following code:

• sensor_decoding.py
• source_decoding.py

Figure 6 is generated using the real data analysis script:

• real_data_faces.py

Dependencies

This work uses MNE-Python and scikit-learn.
It further relies on a library of functions that can be found under: https://github.com/britta-wstnr/python_code_base
For visualization, NiBabel and Matplotlib are used.