DataInt.get_activation_outhead

DataInt.get_activation_outhead() → str | torch.nn.Module | None

Tries to interfere the activation function for the output head, based on the last transformation. It follows the following rules:

• If the last transformation is a MinMaxScaler with a target range of (0, 1), the activation function is sigmoid.

• If the last transformation is a MinMaxScaler with a target range of (-1, 1), the activation function is tanh.

• Otherwise, no activation function is inferred. None is returned.

Returns:

Optional[Union[str, nn.Module]] – Activation function for the output head.