Many neural network packages and applications of neural networks treat the models as black boxes. That is, people tend to utilize some framework that allows them to quickly build a neural network using a bunch of default values and automation. They are often able to produce some results, but this sort of convenience usually does not build much intuition about how the models actually work. As a result, when the models do not behave as expected, it's very hard to understand why they might be making weird predictions or having trouble converging.
Before jumping into more complicated neural networks, let's build up some basic intuition about neural networks such that we do not fall into this pattern. We are going to build a simple neural network from scratch to learn about the basic components of neural networks and how they operate together...
