simple_perceptron.py

#!/usr/bin/env python
# -*- coding: utf-8 -*-

import numpy as np
import matplotlib.pyplot as plt
import os.path

class SimplePerceptron:

    def __init__(self, r = 0.5, v = 0):
        self.r = r
        self.v = v

    def step(self, x):
        return int(x >= self.v)

    def add_bius(self, vec):
        return np.append(vec, 1)

    def fit(self, x_vs, y_s):
        self.w_v = np.random.uniform(-1.0, 1.0, (1, len(list(x_vs[0])) + 1))[0]
        while y_s != list(map(self.predict, x_vs)):
            for y, x_v in zip(y_s, x_vs):
                self.w_v = self.w_v + self.r * \
                    (y - self.predict(x_v)) * self.add_bius(x_v)

    def g(self, x_v):
        return np.dot(self.w_v, x_v)

    def predict(self, x_v):
        x_v = self.add_bius(x_v)
        return self.step(self.g(x_v))


def plot(sp, x_vs, y_vs, fname):
    """plot function
    """
    plt.cla()
    plt.clf()
    plt.close()

    plt.ylim((-1,2))
    plt.xlim((-1,2))

    # plot points
    for i in range(len(x_vs)):
        x_v = x_vs[i]
        x = x_v[0]
        y = x_v[1]

        if y_vs[i] == 1:
            plt.plot(x, y, "r.", label= "["+ str(x) + ", "+ str(y) + "] class 1")
        else:
            plt.plot(x, y, "k.", label= "["+ str(x) + ", "+ str(y) + "] class 2")

    xs = range(-5, 5)
    ys = [-(sp.w_v[0] / sp.w_v[1]) * x - (sp.w_v[2] / sp.w_v[1])
          for x in xs]
    path = os.path.join("graphs", fname)

    plt.plot(xs, ys)
    plt.legend(loc="best")
    plt.savefig(path)


def main():

    # 教師データ
    X_train = [[1, 1], [1, 0], [0, 1], [0, 0]]
    y_train = [1, 0, 0, 0]  # AND
    sp1 = SimplePerceptron()
    sp1.fit(X_train, y_train)
    plot(sp1, X_train, y_train, "graph_and.png")

    y_train = [1, 1, 1, 0]  # OR
    sp2 = SimplePerceptron()
    sp2.fit(X_train, y_train)
    plot(sp2, X_train, y_train, "graph_or.png")


if __name__ == "__main__":
    main()