search

GriffithsLab / whobpyt

Whole-Brain Modelling in PyTorch
https://griffithslab.github.io/whobpyt
MIT License
26 stars 22 forks source link

Transparency of Model Parameters #108

Open Andrew-Clappison opened 1 year ago

Andrew-Clappison commented 1 year ago

Convert RWW code to use log value so that the parameter returned by par() is actual value used (instead of min + ReLU(stored_value) being used). Convert Jansen-Rit to use log value so that the parameter returned by par() is actual value used (instead of min + ReLU(stored_value) being used).

Andrew-Clappison commented 11 months ago

To be done at a later time.

Davi1990 commented 10 months ago

save it for the new year @Davi1990 happy to help in case is necessary

Considering using the exponential (instead of ReLu) for preventing to go negative

Andrew-Clappison commented 9 months ago

New approach (please assign someone else):

See if the minimum values above 0 can be removed or replaced using the asLog features, instead of adding a new ReLu feature to the par class.

Locations around: https://github.com/GriffithsLab/whobpyt/blob/main/whobpyt/models/RWW/wong_wang.py#L361 https://github.com/GriffithsLab/whobpyt/blob/main/whobpyt/models/JansenRit/jansen_rit.py#L295