search

coarse-graining / cgnet

learning coarse-grained force fields
BSD 3-Clause "New" or "Revised" License
57 stars 26 forks source link

RBF normalization #176

Closed nec4 closed 4 years ago

nec4 commented 4 years ago

Development:

Heyo! This is a small functionality expansion to the new RBF layers, allowing for the optional normalization of RBF layer outputs. This is controlled by a boolean kwarg, normalize_output in the __init__ methods of the GaussianRBF and PolynomialCutoffRBF. If this option is set to True, the output of the that layer will be divided by the scalar sum of all the outputs of each basis function. By default, this option is set to False - I decided this default behavior because we have not normalized any RBF outputs in any of our models up to this point. Of course, if there is a better reason to make it True by default, we can discuss. Let me know if you see anything amiss or if you would like to add/change functionality!

nec4 commented 4 years ago

I think thats all I wanted to add here! If you are happy with it, then I am happy for a merge.

brookehus commented 4 years ago

Amazing! In the future feel free to indicate by changing the label to "merge candidate to master"!