This project was originally designed for a Kaggle competition in which data was given about the state of a game of DotA (a popular computer game) and the task was to predict which team would be more likely to win. However, it eventually evolved into an attempt to learn how to construct neural networks in general.
Unfortunately, for unknown reasons the network always predicted that the first team won, and the competition ended before I figured out what went wrong.
The main result of this project; a class-based implementation of a backpropogation neural network with one hidden layer.
A Jupyter Notebook which I used to develop NeuralNetwork.py.
A folder containing csv files which describe the match state and the winners of each match.
Various early iterations of the code used for the competition.