perceptron.py

from collections import namedtuple


Input = namedtuple('Input', ['index', 'value'])

def identity(x):
    return x

def step(x):
    return 1 if x >= 0 else -1


class OutputNeuron:

    def __init__(self, weights, fn=identity):
        self.order = len(weights)
        self.weights = weights
        self.inputs = [0] * self.order
        self.fn = fn

    def activate(self, input):
        self.inputs[input.index] = input.value
        return self.activate_next()

    def activate_many(self, inputs):
        for i, value in inputs:
            self.inputs[i] = value
        return self.activate_next()

    def activate_all(self, inputs):
        self.inputs = list(inputs)
        return self.activate_next()

    def activate_next(self):
        return self.compute()

    def compute(self):
        return self.fn(sum(map(lambda x,y: x*y, self.weights, self.inputs)))

    def train(self, table, h):
        print('Train')

        error = False
        for inputs, output in table.items():
            y = self.activate_all(inputs)
            print(self.weights, inputs, y, output)
            if y != output:
                error = True
                for i in range(self.order):
                    self.weights[i] += h * inputs[i] * output

        if error:
            self.train(table, h)

    def __repr__(self):
        return 'Neuron I:{} O:{}'.format(self.inputs, self.compute())


class Neuron(OutputNeuron):

    def __init__(self, weights, synapsis, fn=identity):
        super().__init__(weights, fn)
        self.output = synapsis

    def activate_next(self):
        input = Input(self.output.index, self.compute())
        return self.output.neuron.activate(input)


if __name__ == '__main__':
    n = OutputNeuron([0, 0, 0], step)
    print(n)
    table = {
        (1, -1, -1): -1,
        (1, -1, 1): -1,
        (1, 1, -1): -1,
        (1, 1, 1): 1
    }
    n.train(table, 0.5)
    print(n)