search

main-educational / brain_encoding_decoding

A jupyter book on brain decoding using functional magnetic resonance imaging
https://main-educational.github.io/brain_encoding_decoding/intro.html
MIT License
8 stars 10 forks source link

GCN parameter `n_timepoints` cannot be any value that's not 1 #9

Closed htwangtw closed 2 years ago

htwangtw commented 2 years ago

The other parameters of each layer were hard coded. For example, 675 is the number of voxels in the ROI mask for subject 4 I will need to look into it to find out how to allow different number of TRs, and possibly for other parameters for different subjects. cc @srastegarnia

It will be useful to know how these parameters were set: https://github.com/main-educational/brain_encoding_decoding/blob/6fbd95c3d93e5e7d82628f901c60fa813bf234bb/src/gcn_model.py#L13-L18

If we set n_timewindow=3, we got this error:

RuntimeError: mat1 and mat2 shapes cannot be multiplied (10800x1 and 3x32)
htwangtw commented 2 years ago

I have resolved the issue - the window length in data preprocessing has to match the parameter used in the model Will think about how the tutorial flow to make this clearer.