brain

This project is an implementation of artifical neural networks and the backpropagation algorithm from scratch.

Example

from neuralnetwork import Brain

# Network with 3 inputs, two hidden layers with 6 and 4 neurons respecively 
# and 2 output neurons. With hyperbolic tangent as activation function.
brain = Brain([3, 6, 4, 2], "Hyp tan")

# xs <- input data
# ys <- corresponding output data

xs = [
    [2.0, 3.0, -1.0],
    [2.0, -1.0, 0.5],
    [0.5, 1.0, 1.0],
    [1.0, 1.0, -1.0],
    [2.0, 2.0, 0.7]
]

ys = [[1, 0], [0, 1], [0, 1], [1, 0], [1, 0]]

# Learn for 2000 epochs with a learn rate of 0.01
brain.learn(2000, 0.01, xs, ys)

Training larger networks

The nextwork can learn to classify handwritten digits. In handwrittendigits.py this is done on the MNIST dataset with approximately 90% accuracy on unseen data.

For more visual examples: circle.py and halfmoons.py train the network to classify points in 2d space with the following decision boundaries: