CondVAEModel.evaluate

CondVAEModel.evaluate(data: DataModule | Tuple[torch.Tensor, torch.Tensor] | Tuple[ndarray, ndarray], untransform: bool = False, use_objective: bool = False, losses_evaluate: Dict[str, torch.nn.Module] | None = None) → Tuple[DataFrame, DataFrame, DataFrame, DataFrame][source]

Evaluate the model on the validation data. This method is similar to CondAEModel.validate(), but allows to compute the losses in the untransformed space.

Depending on the argument untransform, the losses computation is applied on the transformed or untransformed (and non post-processed) data. Hence, one has to be careful when providing the losses to be evaluated, as they should be consistent with the data space - or better work in both spaces.

Parameters:

  • data (Union[DataModule, Tuple[torch.Tensor, torch.Tensor], Tuple[np.ndarray, np.ndarray]]) – The data module object that provides validation data. Alternatively, a tuple of input and output tensors can be provided.

  • untransform (bool, optional, default=False) – If True, the losses are computed in the untransformed space (i.e., the original space).

  • use_objective (bool, optional, default=False) – If True, the losses are computed using the objective functions of the data objects. Requires untransform=False.

  • losses_evaluate (Optional[Dict[str, nn.Module]], optional, default=None) – A dictionary containing the losses to be evaluated. If None, the losses are computed using loss defined by DataObject.get_loss_evaluation().

Returns:

Tuple[pd.DataFrame, pd.DataFrame, pd.DataFrame, pd.DataFrame] – A tuple containing the evaluation losses for the input and output features, as well as the values predicted for input and output features.