Actually, I added that disclaimer on performance because of your first comment. I realized people were getting the wrong idea about my little example, and were thinking this could be used in place of packages like Caffe, Torch7, Theano, TensorFlow, ect...