Neural Network for Digit Identification 

This project is a highly-configurable Neural Network Implementation for digit verification, complete with test and training data. It uses stochastic gradient descent, and an optional number of layers, layer widths, and activation function. The accuracy for each iteration is provided to the user. After extensive training, the program predicts on a set of testing data and leaves the user to check if the predictions make sense.

Additionally, complicated training measures were taken to avoid overfitting on training data. Stochastic training batches are used to train and validate during each iteration of training the network. The batch size, number of iterations, and gradient descent step size are all configurable in this implementation

The data is interpreted as integer values for each pixel in the number image, and it is the neural networks job to train on these values and learn through the use of gradient descent to readjust the weight vectors across all the neurons.

This code is completely in Python and makes use of extensive prior knowledge about mathematics and statistics in order for the implementation to have been done. I am responsible for the development the entire project and have learned a lot during the month that I worked on this. 

The result of this project was a huge accomplishment in learning about machine learning and implementing the complicated mathematics and statistical concepts manually in a tried and true implementation neural networks. This subject is a huge passion of mine and it was very valuable and fun for me to experience!