perceptron neural network

perceptron.py

#!/usr/bin/env python
# -*- coding: utf-8 -*-
"""
Perceptron neural network.  Can learn any linearly separable logic.

(c) Jack Peterson, 6/12/2008
"""
from random import random

def hardlimit(row, col, bias):
    """
    Hard limit evaluation function.
    
    If dot product of the state and input vectors is greater than or equal
    to the bias, then neuron fires.
    """
    product = 0
    for r, c in zip(row, col):
        product += r * c
    if product >= bias:
        fire = 1
    else:
        fire = 0
    return fire

def transition(state, invec, outvec, goal, bias):
    """
    Transition function.

    Compare the binary values of the current output element to the goal
    element, and adjust the state vector if different.
    """
    # For every output element that is not equal to its corresponding goal
    # element, adjust the state vector and the bias
    num_states = len(state)
    if outvec != goal:
        for i in xrange(num_states):
            state[i] += (goal - outvec) * invec[i]
        bias -= goal - outvec
    return state, bias

def main():
    invec = [[0, 0], [0, 1], [1, 0], [1, 1]]

    # Randomize on (0, 1] both state vector elements
    state = [random(), random()]

    # Randomize on (0, 1] bias value
    bias = random()

    # Initial output list
    outvec = [0, 0, 0, 0]

    # Desired output list: simple OR
    # (perceptron can learn any linearly separable set)
    goal = [0, 1, 1, 1]

    print 'Simple OR:', goal
    print '----------------------------'
    print 'Output         State    Bias'
    print '----------------------------'
    elements = len(invec)
    for j in xrange(100):

        # Iterate the 4 input vectors through the evaluation function
        for i in xrange(elements):
            outvec[i] = hardlimit(state, invec[i], bias)

        # If the output vector has reached the goal vector, then quit
        if outvec == goal:
            print outvec, ' ', state, ' ', bias
            print 'Training complete!'
            return

        # Iterate the output values through the transition function, and
        # retrieve the adjusted values for the state vector and the bias
        for i in xrange(elements):
            state, bias = transition(state, invec[i], outvec[i], goal[i], bias)
            print outvec, ' ', state, ' ', bias

if __name__ == "__main__":
    main()