MEG (and possibly later fMRI) analysis scripts for study group at Adv. Cognitive Neuroscience, Aarhus University.

For each group:

• Expects two numpy .npy files
  • trials: with shape (trials, sensors, time points).
  • labels: with shape (trials)
• This file should be put in data/group_x/.
• Then you run precompute_time_point_data.py (change variables first).
• Now you can run either single_time_point_analysis.py, for testing the code on a single time point, or all_time_points_analysis.py, for repeated cross-validation of each time point. This can take quite a while.
• You can define custom model functions in models.py.
• Results are put in results/time_point_models/single/group_x/.
• You can use plot_time_point_results.R to plot your results. The images will be saved as a .tif file, which you can use directly or convert to png.

While there is code for running leave-one-group-out cross-validation, it is quite slow and doesn't seem to get good results. It's possible that some kind of windowed analysis (e.g. blocks of 10 time points) could work better.

For help: Please ask questions in an issue here on GitHub so everyone can see the answer :)

For reference only, here are the plots from our results: