Remove additional softmax function if ClassificationNeuralNetwork uses CategoricalCrossEntropy

[Marsmaennchen221]

Describe the bug

Currently, the last layer of a ClassificationNeuralNetwork module uses softmax as an activation function, but this is not required and can lead to problems in the learning because pytorch CategoricalCrossEntropy already includes a softmax function, as described in the documentation:

Note that this case is equivalent to applying LogSoftmax on an input, followed by NLLLoss.

To Reproduce

/

Expected behavior

Don't use an extra softmax function

Screenshots (optional)

No response

Additional Context (optional)

No response