Neural Network

This project attempts to create a multilayer perceptron used to recognize handwritten digits from scratch, using only NumPy and Random Python modules. (currently, the input is limited to that from MNIST dataset only)

Description

There are 3 main methods:

nn.stochastic_gradient_descent(training_data, epoch, batch_size, learning_rate) nn.testing_nn(test_data) nn.eval_input(input_data)

Notes:

• master.py provides example on how the user can run the nn.py module, along with the given data loader

• .stochastic_gradient_descent() returns null.
• .testing_nn() returns 2 int values: the number of correct guess and the total number of trials.
• .eval_input() returns the number represented by the handwritten 28x28 pixels

• training_data is a list of tuples (x, y), where x is an n x 1 NumPy array, where n is the size of the input (In the sample dataset loader, n is 784). As for y, it is a 10 x 1 NumPy array filled with zeros, except for the index of the expected output
• test_data is a list of tuples (x, y) where x is an n x 1 NumPy array, where n is the size of the input. As for y, it is an int value containing the expected output
• input_data is an n x 1 NumPy array (in the example, n is 784)