challenging to get multi-dimensional layer sizes right. torch is not compatible numpy as input, which is painful for pre and post processing. no metrics means n

challenging to get multi-dimensional layer sizes right. torch is not compatible numpy as input, which is painful for pre and post processing. no metrics means no History object. no activation attribute for Linear layers and Softmax layer behaves weird.