梯度下降

原理见ng教程

code

# -*- coding: utf-8 -*-
"""
Created on Tue Sep 01 17:08:36 2015

@author: young
"""
import os
import numpy as np
import matplotlib.pyplot as plt

def getData(filename):
    data = []
    with open(filename) as f:
        alldata = f.readlines()
        print alldata
    for line in alldata:
        line = line.strip('\n')
        x0 = line.split('\t')[0]
        living_area = line.split('\t')[1]
        price = line.split('\t')[2]
        temp = []
        temp.append(x0)
        temp.append(living_area)
        temp.append(price)
        data.append(temp)
    return  data

def genData(data):
    alpha = 0.000005
    theta = np.ones(2)
    X = np.zeros(shape = (len(data),2))
    Y = np.zeros(shape = (len(data)))

    for i in range(0,len(data)):
        X[i][0] = data[i][0]
        X[i][1] = data[i][1]
        Y[i] = data[i][2]

    iteration_num = 10000
    cost = []
    xTrans = X.transpose()
    k = 0
    while k < iteration_num:
        hypothesis = np.dot(X,theta)
        loss = hypothesis - Y
        temp = loss**2
        J = np.sum(temp)
        cost.append(J)
        #calculate the gradient
        gradient = np.dot(xTrans,loss)
        #update
        theta = theta - alpha * gradient
        k = k + 1
    plt.figure(1)
    cx = range(len(cost))
    plt.plot(cx,cost)
    plt.figure(2)
    plt.plot(X[:,1],Y,'b.')
    x = np.arange(0,2,0.1)
    y = x * theta[1] + theta[0]
    plt.plot(x,y)
    plt.margins(0.2)
    plt.show()


if __name__ == '__main__':
    data = getData('ex1.txt')
    genData(data)