perceptron.py

import random,copy
W1 = [[random.random()*.2, random.random()*.2, random.random()*.2]  for i in range(3)]
W2 = [[random.random()*.2, random.random()*.2, random.random()*.2]  for i in range(3)]
#an empty 3x3 matrix
def activation(a_float):
    return round(a_float)

def apply_input(input_in, W):
    input = copy.copy(input_in)
    input.append(-1)
    output = [0.0 for r in range(len(W))]
    for c in range(len(input)):
        for r in range(len(W)):
            output[r] += input[c]*W[r][c]
    return map(activation, output)

def mlp_apply(input_in, layers):
    inputs = [input_in]
    for layer in layers:
        inputs.append(apply_input(inputs[-1], layer))
    return inputs

training = [[[1,0],[1,0,0]], [[1,1], [0,1,0]], [[0,1], [0,0,1]], [[0,0],[0,0,1]]]

eta = .1
def iterate():
    test = random.sample(training, 1)
    output = apply_input(test[0][0], W1)
    deltas = [-1*output[j]+test[0][1][j] for j in range(len(output))]
    if (deltas == [0 for j in range(len(output))]):
        return True
    input = copy.copy(test[0][0])
    input.append(-1)
    for c in range(len(input)):
        for r in range(len(test[0][1])):
            W1[r][c] += eta*deltas[r]*input[c]
    return False

good_enough = False
ohshit = 0
correct_in_a_row = 0
while not good_enough and ohshit < 1000:
    ohshit += 1
    print ohshit
    if iterate():
        correct_in_a_row += 1
        if correct_in_a_row > 12:
            good_enough = True
    else:
        correct_in_a_row = 0