逻辑回归 - matlab与python实现

代码参考了知乎以及机器学习实战的内容，数据集为机器学习实战的testSet.txt

machine learning inaction - Ch05 - testSet.txt

可自行在GitHub搜索获得，此处不提供了。

matlab代码

clear;
clc;
% 逻辑回归
path='E:\Codes\Python\ML\testSet.txt';
[data,label]=loadDataSet(path);
[m,n]=size(data);

alpha=0.001;
theta=ones(n,1);% 3*1
# 循环次数
Loop=10000;

for i=1:Loop
    P = sigmoid(data*theta);
    error=loss(P,label);
    theta=update_theta(alpha,data,error,theta);
end
disp(theta);

%   绘图
%1. 描点，不同分类不同样式
path='E:\Codes\Python\ML\testSet.txt';
dataSet=importdata(path);
[m,n]=size(dataSet); % m*n
x1=[];y1=[];
x2=[];y2=[];
for i=1:m
    if dataSet(i,3)==0
        x1(end+1)=dataSet(i,1);
        y1(end+1)=dataSet(i,2);
    else
        x2(end+1)=dataSet(i,1);
        y2(end+1)=dataSet(i,2);
    end
end

%2. 画线，一条直线分隔两类
xx1=-3.0;
yy1=(-theta(1)-theta(2)*xx1)/theta(3);
xx2=3.0;
yy2=(-theta(1)-theta(2)*xx2)/theta(3);
plot([xx1,yy1],[xx2,yy2]);
hold on
scatter(x1,y1,'k')
hold on
scatter(x2,y2,'r*')


%1. 生成dataMat与labelMat
function [dataMat,labelMat]=loadDataSet(path)
%   dataMat shape为100*3，数据为[1, *, *]酱婶的
%   labelMat shape为100*1，数据为testSet的第三列
testSet=importdata(path);
dataMat=ones(100,1);
Mat1=[1,0,0];
Mat2=[0,1,0;0,0,1;0,0,0];
dataMat=dataMat*Mat1+testSet*Mat2;
labelMat=testSet(:,3);
end

%2. sigmoid函数
function sig=sigmoid(input_x) 
sig=1.0 ./ (1.0+exp(-input_x));
end

%3. 损失函数
function error=loss(P, y)
error=P-y;
end

%4. 更新权值
function theta=update_theta(alpha, X, error, theta)
grad=X'*error;
theta=theta-alpha*grad;
end

python 实现

import pandas as pd
import numpy as np
from matplotlib import pyplot as plt
from numpy.linalg import inv

def loadDataSet(path):
    dataMat = []
    labelMat = []
    fr = open(path)
    for line in fr.readlines():
        lineArr = line.strip().split()         #根据空格切分原来的长字符串
        dataMat.append([1.0, float(lineArr[0]), float(lineArr[1])])#最开始的1.0是最后的常数项
        labelMat.append(int(lineArr[2]))
    fr.close()
    return dataMat,labelMat

# sigmoid函数
def sigmoid(input_x):
    return 1.0/(1.0+np.exp(-input_x))

# 损失函数
def loss(P, y):
    return (P - y)

# 更新权值
def update_theta(alpha, X, error, theta):
    grad = np.dot(X.T, error)
    theta -= alpha * grad
    return theta

# 保存记录目标函数的值，以便观察是否越接近损失theta
def save_cost(P, y, N):
    j = -np.sum(y * np.log(P) + (1 - y) * np.log(1 - P))
    return (1 / N) * j

def plotBestFit(theta):
    n = np.shape(dataMat)[0] #数据个数
    xcord1 = []; ycord1 = [] #第一类数据的坐标
    xcord2 = []; ycord2 = [] #第二类数据的坐标
    for i in range(n):
        if int(labelMat[i]) == 1:
            xcord1.append(dataMat[i,1])
            ycord1.append(dataMat[i,2])
        else:
            xcord2.append(dataMat[i,1])
            ycord2.append(dataMat[i,2])
    fig = plt.figure()
    ax = fig.add_subplot(111)
    ax.scatter(xcord1, ycord1, s=30, c='red', marker='s')
    ax.scatter(xcord2, ycord2, s=30, c='green', marker='*')
    x = np.arange(-3.0, 3.0, 0.1)
    y = (-theta[0] - theta[1]*x)/theta[2]
    ax.plot(x, y)
    plt.xlabel('X1')
    plt.ylabel('X2')
    plt.show()

path='E:\\Codes\\Python\\ML\\testSet.txt'
dataMat, labelMat = loadDataSet(path)
m, n = np.shape(dataMat)
alpha = 0.001
theta = np.ones((n,1)) # 3*1
print(theta)
maxLoop = 300000
dataMat = np.array(dataMat)
labelMat = np.array(labelMat)[:,np.newaxis]

for i in range(maxLoop):
    P = sigmoid(np.dot(dataMat, theta))
    error = loss(P, labelMat)
    theta = update_theta(alpha, dataMat, error, theta)
print("theta = ")    
print(theta)

plotBestFit(theta)