tabensemb.model.AbstractNN.loss_fn

method

AbstractNN.loss_fn(y_pred, y_true, *data, **kwargs)

User defined loss function.

Parameters:

y_true:

The ground truth.

y_pred:

The predictions from the model (from forward() and after before_loss_fn()).

*data:

Tensors of continuous data and derived unstacked data.

**kwargs:

Parameters to train the model returned by AbstractModel._get_params(). It contains all arguments in AbstractModel._initial_values().

Returns:

torch.Tensor

A torch-like loss.

Notes

Other attributes in self can also be used to calculate loss values.