机器学习——KNN算法(K最近邻分类算法)(2020最新版)

素颜马尾好姑娘i 2022-11-26 01:51 194阅读 0赞

### 1、KNN的例子 ###

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L25ld2xpbmZlbmc_size_16_color_FFFFFF_t_70][]

转换为坐标：

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L25ld2xpbmZlbmc_size_16_color_FFFFFF_t_70 1][]

红色是爱情片、蓝色是动作片、黑色是需要判断的点。

**1.1 KNN具体的做法：**

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L25ld2xpbmZlbmc_size_16_color_FFFFFF_t_70 2][]

**其中，欧式距离(欧几里得距离)的计算方法：**

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L25ld2xpbmZlbmc_size_16_color_FFFFFF_t_70 3][]

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L25ld2xpbmZlbmc_size_16_color_FFFFFF_t_70 4][]

### 2、KNN的缺点 ###

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L25ld2xpbmZlbmc_size_16_color_FFFFFF_t_70 5][]

### 3、KNN的实现 ###

# -*- coding: utf-8 -*- #
    """
    -------------------------------------------------------------------------------
      FileName:    knn_algorithm
      Author:      newlinfeng
      Date:        2020/8/2 0002  14:57
      Description: 使用python实现knn算法
    -------------------------------------------------------------------------------
    """
    import matplotlib.pyplot as plt
    import numpy as np
    import operator
    
    #已知分类的数据
    x1 = np.array([3, 2, 1])
    y1 = np.array([104, 100, 81])
    x2 = np.array([101, 99, 98])
    y2 = np.array([10, 5, 2])
    scatter1 = plt.scatter(x1, y1, c='r')
    scatter2 = plt.scatter(x2, y2, c='b')
    
    #未知数据
    x = np.array([18])
    y = np.array([90])
    scatter3 = plt.scatter(x, y, c='k')
    
    #画图例
    plt.legend(handles=[scatter1, scatter2, scatter3], labels=['labelA', 'labelB', 'X'], loc='best')
    plt.show()
    
    
    #已知分类的数据
    x_data = np.array([[3, 104],
                       [2, 100],
                       [1, 81],
                       [101, 10],
                       [99, 5],
                       [81, 2]])
    y_data = np.array(['A', 'A', 'A', 'B', 'B', 'B'])
    x_test = np.array([18, 90])
    
    #计算样本数量
    x_data_size = x_data.shape[0]
    
    #复制x_test,在行的方向将x_test的数据复制了6次
    # np.tile(x_test, (x_data_size, 1))
    
    #计算x_test与每一个样本的差值
    diffMat = np.tile(x_test, (x_data_size, 1)) - x_data
    #对这六组数据再分别进行计算差值的平方
    sqDiffMat = diffMat**2
    #对这六组数据再分别进行求和,axis=1表示对每一行的数据进行求和
    sqDistances = sqDiffMat.sum(axis=1)
    #对这六个数据再分别进行开方
    distances = sqDistances**0.5
    print(distances)
    
    #把数据的索引进行大小的编号
    sortedDistances = distances.argsort()
    
    classCount = {}
    #设置k
    k = 5
    for i in range(k):
        #获取标签
        votelabel = y_data[sortedDistances[i]]
        #统计标签数量
        classCount[votelabel] = classCount.get(votelabel, 0) + 1
    
    #根据operator.itemgetter(1)-第一个值对classCount排序,然后再取倒序
    sortedClassCount = sorted(classCount.items(), key=operator.itemgetter(1), reverse=True)
    
    #获取数量最多的标签
    knnclass = sortedClassCount[0][0]
    print(knnclass)

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L25ld2xpbmZlbmc_size_16_color_FFFFFF_t_70 6][]

![20200810113809967.png][]

### 4、使用KNN做Iris的分类 ###

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L25ld2xpbmZlbmc_size_16_color_FFFFFF_t_70 7][]

**使用knn实现鸢尾花的分类：**

# -*- coding: utf-8 -*- #
    """
    -------------------------------------------------------------------------------
      FileName:    knn_iris
      Author:      newlinfeng
      Date:        2020/8/2 0002  20:58
      Description: 使用knn算法来对iris数据集进行分类
    -------------------------------------------------------------------------------
    """
    #导入算法包以及数据集
    import numpy as np
    from sklearn import datasets
    from sklearn.model_selection import train_test_split
    from sklearn.metrics import classification_report,confusion_matrix
    import operator
    import random
    
    def knn(x_test, x_data, y_data, k):
        #计算样本数量
        x_data_size = x_data.shape[0]
        #复制x_test
        np.tile(x_test, (x_data_size, 1))
        #计算x_test与每一个样本的差值
        diffMat = np.tile(x_test, (x_data_size, 1))-x_data
        #计算差值的平方
        sqDiffMat = diffMat**2
        #求和
        sqDistances = sqDiffMat.sum(axis=1)
        #开方
        distances = sqDistances**0.5
        #从小到大排序
        sortedDistances = distances.argsort()
        classCount = {}
        for i in range(k):
            #获取标签
            votelabel = y_data[sortedDistances[i]]
            #统计标签数量
            classCount[votelabel] = classCount.get(votelabel, 0) + 1
        #根据operator.itemgetter(1)-第1个值对classCount排序，然后再取倒序
        sortedClassCount = sorted(classCount.items(), key=operator.itemgetter(1), reverse=True)
        #获取数量最多的标签
        return sortedClassCount[0][0]
    
    #载入数据
    iris = datasets.load_iris()
    #x_train,x_test,y_train,y_test=train_test_split(iris.data, iris.target.test_size=0.2)#分割数据0.2为测试数据，0.8为训练数据
    
    #打乱数据
    data_size = iris.data.shape[0] #计算一下，iris中一共有多少数据
    index = [i for i in range(data_size)] #得到一个list，从0开始，到data_size-1结束
    random.shuffle(index)
    iris.data = iris.data[index]
    iris.target = iris.target[index]
    
    #切分数据集
    test_size = 40 #设置测试集的个数为40个
    x_train = iris.data[test_size:] #从第40个开始设置为训练集
    x_test = iris.data[:test_size]  #0——40设置为测试集
    y_train = iris.target[test_size:] #标签也是一样
    y_test = iris.target[:test_size]
    
    predictions = []
    for i in range(x_test.shape[0]):
        predictions.append(knn(x_test[i], x_train, y_train, 5))
    
    print(classification_report(y_test, predictions))
    
    print(confusion_matrix(y_test, predictions))

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L25ld2xpbmZlbmc_size_16_color_FFFFFF_t_70 8][]

**2020-08-10 更新**

[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L25ld2xpbmZlbmc_size_16_color_FFFFFF_t_70]: /images/20221124/c7a96b1dbb0e4bd78dc2ff8ac3239170.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L25ld2xpbmZlbmc_size_16_color_FFFFFF_t_70 1]: /images/20221124/7c106fc2d5eb447a9476cca0e3eea6a8.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L25ld2xpbmZlbmc_size_16_color_FFFFFF_t_70 2]: /images/20221124/3ccff620119547569f06b79c304d2e7f.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L25ld2xpbmZlbmc_size_16_color_FFFFFF_t_70 3]: /images/20221124/0b2b5749fd41400ebdfd71f4d1383e83.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L25ld2xpbmZlbmc_size_16_color_FFFFFF_t_70 4]: /images/20221124/f671c011b7e242669400e0fc9508892e.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L25ld2xpbmZlbmc_size_16_color_FFFFFF_t_70 5]: /images/20221124/7437d37686d4443197d8dd55a4164c12.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L25ld2xpbmZlbmc_size_16_color_FFFFFF_t_70 6]: /images/20221124/f0b77a5097714a22b23de6569cf0f4f4.png
[20200810113809967.png]: /images/20221124/3f70e83b5d6a46588b01f449f2c16f6c.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L25ld2xpbmZlbmc_size_16_color_FFFFFF_t_70 7]: /images/20221124/13f6332f3e294064a4328306c6e34556.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L25ld2xpbmZlbmc_size_16_color_FFFFFF_t_70 8]: /images/20221124/afd90c890d694a80afed66ddbf64be17.png