Under the hood -- how do neural networks really work?

Before even calculating the predictions we have to ensure that the data is structured in the same way for the program to process all the different images. Let's make sure that we have two different tensors, each for all of the'nines' and'twos' in the dataset. Before progressing further, we have to discuss what exactly is a tensor. I first heard of the word tensor in the name TensorFlow, which is a (completely!) A PyTorch Tensor in this case is a multidimensional table of data, with all data items of the same type.