ok, will update this above.
looking at your class fakes.py this line: train_ds = TensorDataset(torch.randn(train_length, n_in), torch.randn(train_length, n_out)) and wondering if this way of creating data will allow us to test all functionality (saw a maybe related comment of yours here: Developer chat).
Trying testing the class basic_data / DataBunch / one_item and getting this error
AttributeError: ‘TensorDataset’ object has no attribute 'one_item’
Might be related to the way the data is created - maybe we need to create the fakes stub a little different?