李航《统计学习方法》——用Python2.7实现感知器模型（MNIST数据集）

最近准备入门机器学习，正在看李航老师的《统计学习方法》，刚好看的了博主wds2006sdo写的用Python实现《统计学习方法》里面的算法的文章，深受启发。所以打算在博主wds2006sdo的代码基础上写自己的代码。

代码

基于感知机学习算法的原始形式

#encoding=utf-8  
  
import pandas as pd  
import numpy as np  
import random  
import time  
  
from sklearn.model_selection import train_test_split  
from sklearn.metrics import accuracy_score  
  
def Train(trainset,train_labels):  
    # 获取参数  
    trainset_size = len(train_labels)       #获取数据的总数  
    # 初始化 w,b  
    w = np.zeros((len(trainset[1]),1))      #len(trainset[1]为数据的特征数，w为特征数*1的矩阵  
    b = 0  
  
    study_count = 0                         # 学习次数记录，只有当分类错误时才会增加  
    nochange_count = 0                      # 统计连续分类正确数，当分类错误时归为0  
    nochange_upper_limit = 100000           # 连续分类正确上界，当连续分类超过上界时，认为已训练好，退出训练  
  
    while True:  
        nochange_count += 1  
        if nochange_count > nochange_upper_limit:  
            break  
  
        # 随机选的数据  
        index = random.randint(0,trainset_size-1)  
        img = trainset[index]                #有784个属性的属性集  
        label = train_labels[index]          #结果集  
  
  
        # 计算yi(w*xi+b)  
        yi = int(label != object_num) * 2 - 1                       # 如果等于0, yi= -1, 否则yi=1  
        result = yi * (np.dot(img,w) + b)  
  
        # 如果yi(w*xi+b) <= 0 则更新 w 与 b 的值  
        if result <= 0:  
            img = np.reshape(trainset[index],(len(trainset[1]),1))    # 将img转置，使之成为竖向的矩阵，与w保持规格统一  
  
            w += img*yi*study_step                                  # 按算法步骤3更新参数  
            b += yi*study_step  
  
            study_count += 1  
            if study_count > study_total:  
                break  
            nochange_count = 0  
  
    return w,b  
  
def Predict(testset,w,b ):  
    predict = []  
    for img in testset:  
        result = np.dot(img,w) + b  
        result = result > 0  
  
        predict.append(result)  
  
    return np.array(predict)  
  
study_step = 0.0001                                 # 学习步长  
study_total = 10000                                 # 学习次数  
object_num = 0                                      # 分类的数字  
  
  
  
if __name__ == '__main__':  
  
    print 'Start read data'  
  
    time_1 = time.time()  
  
    raw_data = pd.read_csv('train_binary.csv',header=0)  
    data = raw_data.values  
    imgs = data[0::,1::]  
    labels = data[::,0]  
  
    # features = get_hog_features(imgs)  
    features = imgs  
    # 选取 2/3 数据作为训练集， 1/3 数据作为测试集  
    train_features, test_features, train_labels, test_labels = train_test_split(features, labels, test_size=0.33, random_state=23323)  
  
    time_2 = time.time()  
    print 'read data cost ',time_2 - time_1,' second','\n'  
  
    print 'Start training'  
    w,b = Train(train_features,train_labels)  
    time_3 = time.time()  
    print 'training cost ',time_3 - time_2,' second','\n'  
  
    print 'Start predicting'  
    test_predict = Predict(test_features,w,b)  
    time_4 = time.time()  
    print 'predicting cost ',time_4 - time_3,' second','\n'  
  
    score = accuracy_score(test_labels,test_predict)  
    print "The accruacy socre is ", score