search

kutaslab / fitgrid

Multichannel event-related time-series regression modeling for EEG, MEG, and sensor array data
https://kutaslab.github.io/fitgrid
BSD 3-Clause "New" or "Revised" License
8 stars 4 forks source link

[JOSS] Manuscript - alternative mass univariate model fits #205

Closed AJQuinn closed 3 years ago

AJQuinn commented 3 years ago

Hello - the JOSS manuscript is clear and informative. My only comment woulld be to acknowlege existing python tools which can fit mass-univariate regression models.

The combinatorial explosion highlighted in the statement of need (paragraph starting on line 36) is a critical point but several existing tools already allow hundreds or thousands of models to be estimated simultaneously. For example np.linalg.lstsq and sklearn.linear_model.LinearRegression both repeat a regression across many 'targets'. The user would only have to reshape the data array to be of shape [n_trials x n_samples*n_channels] before calling the fit function. fitgrid adds a lot of worthwhile functionality beyond this point (ephys specific support, mixed-models etc) - but worth acknowledging that other python tools can contain the combinatorial explosion for a standard regression problem.

AJQuinn commented 3 years ago

Thanks for the update - happy to consider this one closed.