Currently, I have a Dataloader initialized with a dataset of an array of tensors that does not inherit from Dataset. So far my model has not been learning at all. I was wondering if it is necessary to pass in a dataset class that inherits from Dataset for Dataloader to work or not.
I don’t think it should really matter unless your dataloader is missing some functions (which should throw an error). If the data and label tensors returned by your dataset are reasonable it is likely the problem is somewhere else.
thanks for the awesome information.