As Lorenzo asked for the performance curves to be of the unscaled loss, the retrain_regressor and model train_step and validation_step needed to be adjusted to return these values.
Also included is the (very messy) jupyter notebook I was using to generate some of the plots previously shown
As Lorenzo asked for the performance curves to be of the unscaled loss, the
retrain_regressorand modeltrain_stepandvalidation_stepneeded to be adjusted to return these values.Also included is the (very messy) jupyter notebook I was using to generate some of the plots previously shown