This PR adds the ability to load separate transforms for input/output/hybrid variable sets (or fit different standard normalization transforms if not specified). These are organized in a new TransformerGroup class that replaces the single autoencoder attribute of the reservoir models.
If the output variable list is different from the input variable list, the training script will use the output transformer when preparing the paired output set.
Added public API:
The previous single autoencoder_path parameter in the ReservoirTrainingConfig is now replaced by a TransformerConfig which contains optional paths for input, output, and hybrid transformers
Refactored public API:
Aforementioned API change breaks old configs
Breaks old model loading, since models now dump and load a TransformerGroup instead of a single autoencoder
This PR adds the ability to load separate transforms for input/output/hybrid variable sets (or fit different standard normalization transforms if not specified). These are organized in a new
TransformerGroupclass that replaces the singleautoencoderattribute of the reservoir models.If the output variable list is different from the input variable list, the training script will use the output transformer when preparing the paired output set.
Added public API:
autoencoder_pathparameter in theReservoirTrainingConfigis now replaced by aTransformerConfigwhich contains optional paths for input, output, and hybrid transformersRefactored public API:
Aforementioned API change breaks old configs
Breaks old model loading, since models now dump and load a
TransformerGroupinstead of a single autoencoder[ x] Tests added