HaskelANN

A feed-forward Artificial Neural Network implementation in Haskell with a stochastic gradient descent backpropagation training algorithm. Gradients are calculated using an automatic differentiation implementation.

Instructions

The ANN framework is fairly self contained, there are only 4 source files required to run the two demonstrations:

• ANN.hs - the main ANN framework
• AutomaticDifferentiation.hs - provides automatic differentiation types
• DataProcessing.hs - functions for loading and preparing data for machine learning
• Demo.hs - two classification demonstrations (requires iris.csv and breast_cancer.csv - see below)

To run the demonstrations:

1. Open GHCI
2. :load Demo
3. Call either the 'demo1' or 'demo2' function

Data

The demonstrations use freely available data from the UCI Machine Learning repository.