Analyzing EEG / MEG data with PyMC3

Questions
2

Hi @thht

if you don’t mind the opinion from a PyMC3 novice here are some ideas

  1. You might want to consider a more informative prior for the scale parameter, at the moment is extremely wide. I usually go for an HalfCauchy distribution. You can center it on a very small value if you are afraid that large values can cause problems.
  2. Given the nature of EEG signal (if I recall correctly from my master) you might want to model the correlation between electrodes.
  3. It might be an overkill here, but PyMC3 offers approximate variational inference which can provide consistent speed-up.

Finally,
can you expand a bit on the nature of n_samples? I struggle to depict the experimental design (I was expecting a (n_participants, n_channels) matrix).

1 Like