`swissrivernetwork.benchmark.model`¶

swissrivernetwork.benchmark.model ¶

Bases: Module

Bases: TransformerEmbeddingModel

ExtrapoLstmModel(
    *args, extrapo_mode: str | None = None, **kwargs
)

Factory function to create LSTM model for timewise extrapolation.

Parameters:

Name	Type	Description	Default
`extrapo_mode`	`str \| None`	str \| None Extrapolation mode. Options are: - 'limo': Last Input Multiple Output, use the last input to predict multiple future steps directly. - 'future_embedding': Use learnable future step embeddings to fill in future steps. Then use standard LSTM to predict all steps together. - 'recursive': Recursive prediction, predict one step at a time and feed it back as input for the next step. (not implemented yet) - None: Default setting (LIMO).	`None`