Enabling generator pretraining for WGAN
To stabilize training when downscaling EURAD-IM forecasts in scope of the DestinE-AQ DE370c-use case, pretraining of the WGAN's generator model will be tested.
This requires an adaption of the WGAN-architecture, its hyperparameters as well as its customized train_step
-method.
Edited by Michael Langguth