The evolving number of models for hysteresis modeling require consolidation and standardization of validation and test metrics in order to fairly compare models.
This goes hand-in-hand with Define a fixed training and validation set.
In principle we should use MAE, MSE, MAPE, SMAPE.
Outcome
Implemented with Transformers for Time Series v0.0.18 https://gitlab.cern.ch/dsb/hysteresis/transformertf/-/tree/v0.0.18?ref_type=tags which implements MAPE and SMAPE in transformertf.nn.functional module. MSE and MAE losses are already defined in torch.