Linear Unit of neural network with numpy

np-ai.py

#!/usr/local/bin/python
# -*- coding: UTF-8 -*-
import numpy as np
import matplotlib.pyplot as plt


def predict(input_vec, f):
    # print (input_vec, W)
    return f(np.dot(input_vec, W))


def train(input_vecs, labels, iterations, rate, f):
    for _ in xrange(iterations):
        one_iteration(input_vecs, labels, rate, f)

def one_iteration(input_vecs, labels, rate, f):
    samples = zip(input_vecs, labels)
    for (input_vec, label) in samples:
        input_vec = np.append(input_vec, [1])
        output = predict(input_vec, f)
        update_W(input_vec, output, label, rate)

def update_W(input_vec, output, label, rate):
    delta = label - output
    global W
    W = W + rate * delta * input_vec

if __name__ == '__main__':
    input_vecs = [[1, 1], [0, 0], [1, 0], [0, 1]]
    # 期望的输出列表，注意要与输入一一对应
    # [1,1] -> 1, [0,0] -> 0, [1,0] -> 0, [0,1] -> 0
    labels = np.array([1, 0, 0, 0])
    # 形如w2x2+w1x1+bias, 或w2x2+w1x1+w0
    input_num = 2 
    W = np.array([0.0] * (input_num+1))

    train(input_vecs, labels, iterations=10, rate=0.1, f=lambda x: 1 if x > 0 else 0
)
    print W

    for x in input_vecs:
        print 'AND(%s) = %d' % (x, predict(x+[1], f=lambda x: 1 if x > 0 else 0
))

    ###############
    # 形如w1x1+bias, 或w1x1+w0
    input_vecs = [[5], [3], [8], [1.4], [10.1]]
    # 期望的输出列表，月薪，注意要与输入一一对应
    labels = [5500, 2300, 7600, 1800, 11400]    
    input_num = 1
    W = np.array([0.0] * (input_num+1))

    train(input_vecs, labels, iterations=10, rate=0.01, f=lambda x: x)
    print W

    for x in [[3.4], [15], [1.5], [6.3]]:
        t = predict(x+[1], f=lambda x: x)
        print 'Work %s years, monthly salary = %.2f' % (x, t)

    plt.figure(1)
    z = np.linspace(0, 12, 100)
    plt.plot(z, z * W[0] + W[1])
    for (i, l) in zip(input_vecs, labels):
        plt.plot(i, l, 'g^')    
    plt.show()