NeuralNetBinaryClassifier was not working with torch.compile because the non-linearity was not correctly inferred. This inference depends on the instance type of the criterion. However, when using torch.compile, the criterion is wrapped, resulting in the isinstance check to miss. Now, we unwrap the criterion before checking the instance type.
Fixes #1057
NeuralNetBinaryClassifierwas not working withtorch.compilebecause the non-linearity was not correctly inferred. This inference depends on the instance type of the criterion. However, when usingtorch.compile, the criterion is wrapped, resulting in theisinstancecheck to miss. Now, we unwrap the criterion before checking the instance type.