theseus.Objective.error

Objective.error(input_tensors: Optional[Dict[str, Tensor]] = None, also_update: bool = False) → Tensor

Evaluates the error vector.

Parameters

• input_tensors (Dict[str, torch.Tensor], optional) – if given, it must be a dictionary mapping variable names to tensors; if a variable with the given name is registered in the objective, its tensor will be replaced with the one in the dictionary (possibly permanently, depending on the value of also_update). Defaults to None, in which case the error is evaluated using the current tensors stored in all registered variables.

• also_update (bool, optional) – if True, and input_tensors is given, the modified variables are permanently updated with the given tensors. Defaults to False, in which case the variables are reverted to the previous tensors after the error is evaluated.

Returns

a tensor of shape (batch_size x error_dim), with the

concatenation of all cost functions error vectors. The order corresponds to the order in which cost functions were added to the objective.

Return type

torch.Tensor