ENH: More features for ReceptiveField

larsoner commented 7 years ago

[x] support score method for use with GridSearchCV
[x] use custom Ridge solver for speed and multiple regularization types (i.e., Ross's Fourier-based code)
[x] change to positive lags for causal behavior #4550
[x] backward transform
[x] Add CUDA support for the fitting
[ ] if necessary, make tmin/tmax/sfreq logic consistent with Epochs class
[ ] forward/backward model tutorial?
[ ] plot method that is smart enough to at least deal with 1D (STA) and 2D (STRF) plotting, probably using NonUniformImage class rather than pcolormesh
[ ] decimation (decimate data before fitting or model after fitting?)
[ ] refactor with linear_regression_raw, XDawn (categorical data, #4940)

larsoner commented 6 years ago

... oh wait, looks like the patterns_ part is already done!

nbara commented 6 years ago

:) yup ! I even added a small section in the tutorial for this

I haven't used RF much lately but one feature I remember missing is the ability to subsample the lags/delays, or to provide a list of lags instead of tmin and tmax. Of course you can always downsample your data but it would be more convenient to have this feature directly in ReceptiveField.

mne-tools / mne-python

ENH: More features for ReceptiveField #3884