ZachElkins / BasicNeuralNetwork

Basic Neural Network

A simple python implementation of a neural network for learning about the subject.

The project is made up of five important files

• train.py — Create and train a neural network
• test.py — Test the error rate of a neural network created with train.py
• neuralnetwork.py — The NeuralNetwork class implementation.
• activation_functions.py — A dict with the different activation functions, and their derivatives, that can be used in the nueral networks.
• utils.py — Miscellaneous functions utilized by other processes.

Train.py and Test.py have optional command line arguments listed below

train.py

usage: train.py [-h] [-hn HN] [-a A] [-lr LR] [-t T] [-g] [-n N]

Train a Neural Network on the mnist dataset.

optional arguments:
  -h, --help  show this help message and exit
  -hn HN      The number of hidden nodes to use. (Default: 700)
  -a A        The activation function to use. Sigmoid, Tanh, ReLU. (Default: Sigmoid)
  -lr LR      The learning rate. Must be within the exclusive range (0, 1). (Default: 0.05)
  -t T        The number of images used to train the model in the range (0, 60000]. (Default: 30000)
  -g          Flag to graph the training over time. (This will add significant time to training)
  -n N        Filename, not including extension. (Default: model)

test.py

usage: test.py [-h] [-t T] [-f F]

Test a Neural Network created with train.py

optional arguments:
  -h, --help  show this help message and exit
  -t T        The number of images used to test the model in the range (0, 10000]. (Default: 1000)
  -f F        Filename of model to test, not including extension. (Default: model)