Nrel-phygnn

What’s Changed

Significant refactor to allow for convolutional layers in the phygnn model along with multi dimensional features or labels (tested up to 5D features). Includes BREAKING CHANGES with the phygnn model save/load functionality. The model Layers handler object is now saved directly to the pkl file instead of trying to save and load model weights.

* Gb/conv rnn layers (24) grantbuster
* Gb/nsrdb gan (23) grantbuster

What’s Changed

* Gb/better history (22) grantbuster

What’s Changed

* incremented veresion for doi release (21) grantbuster

What’s Changed

Breaking change: switched position of y_true and y_predicted in p_fun to match the keras metrics methods.

* Gb/gan (20) grantbuster

Bug Fixes

* add shuffle to model interfaces (19) MRossol

What’s Changed

phygnn wasnt executing dropout layers correctly. The Dropout layer object needed a boolean flag to know if it was training or not. Fixed with additional tests.

* Gb/dropout bug (18) grantbuster

What’s Changed

* phygnn save() to now save optimizer config dict instead of optimizer … (17) grantbuster