tabensemb.model.AbstractNN._forward#

method

AbstractNN._forward(x: Tensor, derived_tensors: Dict[str, Tensor]) Tensor[source]#

The real forward method.

Parameters:
x

A tensor that contains continuous features.

derived_tensors

It mostly has the same structure as the derived unstacked data derived_data stored in a tabensemb.data.datamodule.DataModule. If some models are required (defined in AbstractModel.required_models()), there will be a key named “data_required_models” containing the data batch, the output, and possibly the hidden representation of the required models.

Returns:
torch.Tensor

The output of the model.

Notes

For classification tasks, DO NOT turn logits into probabilities here because we have already implemented this later. See also output_norm().