机器学习实战篇 (k近邻算法)

k近邻算法：通过测量不同特征值之间的距离进行分类

优点:精度高，对异常值不敏感，无数据输入假定。

缺点：计算复杂度高，空间复杂度高。

计算公式

分类器的代码实现

import numpy as np
from collections import Counter

def classify0(inx, dataset, labels, k=1):
    ##预处理(此处的输入labels是带有具体分类内容的list),inx和dataset都numpy对象
    if k <= 0:
        k = 1
    try:
        y = inx.shape[1]
    except:
        inx.shape=(-1, inx.shape[0])
    ##计算欧氏距离
    num_test = inx.shape[0]
    num_train = dataset.shape[0]
    dists = np.zeros((num_test, num_train))
    dists = np.multiply(np.dot(inx, dataset.T), -2)
    inx_sq = np.sum(np.square(inx), axis=1, keepdims=True)
    dataset_sq = np.sum(np.square(dataset), axis=1)
    dists = np.add(dists, inx_sq)
    dists = np.add(dists, dataset_sq)
    dists = np.sqrt(dists)
    ###获取标签
    result = []
    per_line_labels=[]
    sort_arg = dists.argsort()[:,:k]
    for line in sort_arg:
        per_line_labels = [labels[index] for index in line]
        result.append(Counter(per_line_labels).most_common(1)[0][0])
    return result

实例1 利用K-近邻算法改进约会网站的配对效果

数据集下载 http://pan.baidu.com/s/1geMv2mf

1.从文件中读取数据转化为可计算的numpy对象

def file1matrix(filename):
    ###从文件中读取数据并转为可计算的numpy对象
    dataset = []
    labels = []
    with open(filename,'r') as f:
        for line in f:
            line = line.strip().split('\t')
            labels.append(line.pop())
            dataset.append(line)
    dataset = np.array(dataset, dtype=np.float32)
    return dataset, labels

2.将数据可视化

def convert(labels):
    label_names = list(set(labels))
    labels = [label_names.index(label) for label in labels]
    return label_names,labels

def draw(dataset, labels, label_names):
    labels = [ i+1 for i in labels]  ###下标加1，绘色
    from matplotlib import pyplot as plt
    from matplotlib import font_manager
    zhfont = font_manager.FontProperties(fname='C:\\Windows\\Fonts\\msyh.ttc')
    plt.figure(figsize=(8, 5), dpi=80)
    ax = plt.subplot(111)
    # ax.scatter(dataset[:,1], dataset[:,2], 15.0*np.array(labels), 15.0*np.array(labels))
    # plt.show()
    type1_x = []
    type1_y = []
    type2_x = []
    type2_y = []
    type3_x = []
    type3_y = []
    for i in xrange(len(labels)):
        if labels[i] == 1:
            type1_x.append(dataset[i][0])
            type1_y.append(dataset[i][1])
        if labels[i] == 2:
            type2_x.append(dataset[i][0])
            type2_y.append(dataset[i][1])
        if labels[i] == 3:
            type3_x.append(dataset[i][0])
            type3_y.append(dataset[i][1])
    ax.scatter(type1_x, type1_y, color = 'red', s = 20)
    ax.scatter(type2_x, type2_y, color = 'green', s = 20)    
    ax.scatter(type3_x, type3_y, color = 'blue', s = 20)    
    plt.xlabel(u'飞行里程数', fontproperties=zhfont)
    plt.ylabel(u'视频游戏消耗时间', fontproperties=zhfont)
    ax.legend((label_names[0], label_names[1], label_names[2]), loc=2, prop=zhfont)
    plt.show()

3.归一化特征值 (这里介绍两种方法)

####由于数据中飞行里程数特征值与其他的特征值差距较大，对计算结果会产生非常大的影响，所以将特征值转化为0到1区间内的值   
def autoNorm0(dataset):
    if not isinstance(dataset, np.ndarray):
        dataset = np.array(dataset,dtype=np.float32)
    ###归一化特征值 newvalue = (oldvalue - min) / (max - min)
    minVals = dataset.min(0)
    maxVals = dataset.max(0)
    ranges = maxVals - minVals
    dataset = dataset - minVals
    dataset = dataset / ranges
    return dataset

def autoNorm1(dataset):
    ###归一化特征值 newvalue = (oldvalue - 均值) / 标准差    ----->推荐使用这种方法
    if not isinstance(dataset, np.ndarray):
        dataset = np.array(dataset,dtype=np.float32)
    mean = dataset.mean(0)
    std = dataset.std(0)
    dataset = dataset - mean
    dataset = dataset / std
    return dataset

4.编写测试代码

def datingTest():
    ##随机选取测试集和训练集
    filename = 'datingTestSet.txt'
    dataset, labels = file1matrix(filename)
    dataset = autoNorm1(dataset)
    train_length = int(dataset.shape[0] * 0.9)
    test_length = dataset.shape[0] - train_length
    from random import sample
    all_index = sample(range(dataset.shape[0]), dataset.shape[0])
    train_index = all_index[:train_length]
    test_index = all_index[-test_length:]
    train_dataset = dataset[train_index, :]
    train_labels = []
    test_dataset = dataset[test_index, :]
    test_labels = []
    for index in train_index:
        train_labels.append(labels[index])
    for index in test_index:
        test_labels.append(labels[index])
    ##训练并计算错误率
    test_result = classify0(test_dataset, train_dataset, train_labels, k=3)
    error = 0
    for res in zip(test_result, test_labels):
        if res[0] != res[1]:
            error += 1
    print 'error accaury:%f' % (float(error) / len(test_labels))

实例2 识别手写数字

1.读取文件数据并转化为可计算的numpy对象

import os

def imgVector(filename):
    vect = []
    with open(filename,'r') as f:
        for line in f:
            line = line.strip()
            vect += [float(n) for n in line]
    number = os.path.split(filename)[-1].split('_')[0]
    return np.array(vect, dtype=np.float32), number

def all_imgVector(directory):
    filelist = os.listdir(directory)
    vects = []
    labels = []
    for filename in filelist:
        vect, label= imgVector(os.path.join(directory, filename))
        vects.append(vect)
        labels.append(label)
    return np.array(vects, dtype=np.float32), labels

2.编写测试代码

def handwritingClassTest():
    test_dir = 'digits\\testDigits'
    train_dir = 'digits\\trainingDigits'
    train_dataset, train_labels = all_imgVector(train_dir)
    test_dataset, test_labels = all_imgVector(test_dir)
    result_labels = classify0(test_dataset, train_dataset, train_labels, k=3)

    error = 0 
    for res in zip(result_labels, test_labels):
        if res[0] != res[1]:
            error += 1
    print 'error accaury:%f' % (float(error) / len(test_labels))