Target parameter required

In BaseWrapper.fit, the targets y are required.

https://github.com/adriangb/scikeras/blob/9f633f5c21a45b43986a8d779c3ef9e3005038bb/scikeras/wrappers.py#L556

This poses challenges for models that don’t require targets (like autoencoders).

[adriangb]

#123 will enable the following behavior:

class AutoEncoderTransformer(BaseWrapper, TransformerMixin):
    def transform(self, X):
        return self.predict(X)

tf = AutoEncoderTransformer(model=...)
encoded = tf.fit_transform(X)

Which is a pretty nice way to bring this behavior into the sklearn ecosystem IMO.

[adriangb]

I was wrong. PR implementing y=None is in #123