Digit-Recognizer

Digit Recognizer Using Machine Learning Built From Scratch. Here I will attempt to explain how it works.

Neural Network

Imagine a neural network a box that takes in an input and returns an output.
Inside this box is a series of layers, the first being the input layer, and the last being the output layer; the ones in-between are hidden layers.
Each layer also has a size which is the amount of neurons it holds (think of each neuron as a node in the network).
Each node is connected to every node in the layer preceding it and following it.
Each connection is weighted and also given a bias (which in a way, can be thought of as an additional weight). This determines how data is processed.
To code this, we need:
- An array containing the sizes of each layer.
- Constants for the size of the input and output of the network, as well as the size of the network.
- A 2D array (layer, neuron) for the outputs of each node.
- A 3D array (layer, neuron, previous neuron) for the weights (since each neuron-to-neuron link has a weight).
- A 2D array (layer, neuron) for the bias of a specific node.
- There is other stuff, but we'll get there eventually.

We generate an array of randomized values as input and pass it to the feed-forward method.
We set the first row of our outputs to the input,
Now we loop through each layer of the network (skipping the input layer), and then loop through each neuron in the layer.
For each neuron, we calculate its output.
- First add the neuron's bias.
- Then sum up the products of the neuron's weight times the outputs for each neuron in the previous layer.
- Finally, put the total sum through a sigmoid function, which transforms any number into a value between 0 and 1.

(Work in Progress)