Closed adam2392 closed 3 years ago
Don’t write this code from scratch
Use existing code for fitting MVAR models
I think the best existing code could be maybe sklearns multivariate linear regression code?
no. sklearn is not a time series package.
you have https://www.statsmodels.org/dev/vector_ar.html and https://github.com/scot-dev/scot/blob/master/scot/var.py
ping @cbrnr
SCoT has all the basic time-varying AR estimation set up. I haven't had time to work on it in the past years, but it is a great package and supports many connectivity measures based on AR coefficients. I'd be happy if we found a way to revive the package, which could even mean transferring it to a new repo and then extending/adapting/updating the code base (it is MIT licensed so there's no problem).
yes +100 to integrate SCoT into mne-connectivity
As mentioned in the original GSoC proposal and also in MNE-Python, this would be a fit to the data using the model:
over small time windows. This is a generative model and so can be accompanied by a tutorial showing "reconstruction" of the ieeg data.