Given any general float torch.Tensor, possibly containing some NaN values, I am looking for an efficient method to either replace all the NaN values in it with zero, or remove them altogether and filter out the "useful" values in another new Tensor.
I am aware that a trivial way to do this is to manually iterate through all the values in the given tensor (and correspondingly replace them with zero or reject them for the new tensor).
Is there some pre-defined Torch function or a combination of functions which can achieve this more efficiently in terms of performance, which relies on the inherent CPU-GPU optimisations of Torch?