[Bug] Standardization of the output and inverse transform of standard deviation about gpytorch

[Balandat]

I am not familiar with StandardScaler but it seems like if that de-means the data that mean would be added to the std in this line:

y_std = y_scaler.inverse_transform(output_std.reshape(-1, 1))

which is not what you want (you only want to scale the std by the scaling factor)

[BidhanChandra]

Thank you for your response. I understood your point. It should be just scaled and the mean of the original values should not be added.
I have another question, does gpytorch have the same facility like normalize_y as in Gaussian process regression implementation of sklearn?

[Balandat]

I don't think GPyTorch has this functionality. BoTorch provides this as a transform though: https://github.com/pytorch/botorch/blob/main/botorch/models/transforms/outcome.py#L223