机器学习理论（一）k近邻算法

点击上方“小白学视觉”，选择加"星标"或“置顶”

重磅干货，第一时间送达

前言

KNN(k-Nearest Neighbors)思想简单，应用的数学知识几乎为0，所以作为机器学习的入门非常实用、可以解释机器学习算法使用过程中的很多细节问题。能够更加完整地刻画机器学习应用的流程。

首先大致介绍一下KNN的思想，假设我们现在有两类数据集，一类是红色的点表示，另一类用蓝色的点表示，这两类点就作为我们的训练数据集，当有一个新的数据绿色的点，那么我们该怎么给这个绿色的点进行分类呢？

一般情况下，我们需要先指定一个k，当一个新的数据集来临时，我们首先计算这个新的数据跟训练集中的每一个数据的距离，一般使用欧氏距离。

然后从中选出距离最近的k个点，这个k一般选取为奇数，方便后面投票决策。在k个点中根据最多的确定新的数据属于哪一类。

KNN基础

1.先创建好数据集x_train, y_train，和一个新的数据x_new, 并使用matplot将其可视化出来。

import numpy as npimport matplotlib.pyplot as plt
raw_data_x = [[3.3935, 2.3313],              [3.1101, 1.7815],              [1.3438, 3.3684],              [3.5823, 4.6792],              [2.2804, 2.8670],              [7.4234, 4.6965],              [5.7451, 3.5340],              [9.1722, 2.5111],              [7.7928, 3.4241],              [7.9398, 0.7916]]raw_data_y = [0, 0, 0, 0, 0, 1, 1, 1, 1, 1]x_train = np.array(raw_data_x)y_train = np.array(raw_data_y)
x_new = np.array([8.0936, 3.3657])
plt.scatter(x_train[y_train==0,0], x_train[y_train==0,1], color='g')plt.scatter(x_train[y_train==1,0], x_train[y_train==1,1], color='r')plt.scatter(x_new[0], x_new[1], color='b')plt.show()

1.knn过程

2.计算距离

from math import sqrt


    
distance = []for x in x_train:    d = sqrt(np.sum((x_new - x) ** 2))    distance.append(d)
# 其实上面这些代码用一行就可以搞定# distances = [sqrt(np.sum((x_new - x) ** 2)) for x in x_train]

输出结果：

[10.888422144185997, 11.825242797930196, 15.18734646375067, 11.660703691887552, 12.89974598548359, 12.707715895864213, 9.398411207752083, 15.62480440229573, 12.345673749536719, 14.394770082568183]

将距离进行排序,返回的是排序之后的索引位置

nearsest = np.argsort(distances)

输出结果：array([6, 0, 3, 1, 8, 5, 4, 9, 2, 7], dtype=int64)

取k个点，假设k=5

k = 5topk_y = [y_train[i] for i in nearest[:k]]topk_y

输出结果：[1, 0, 0, 0, 1]

根据输出结果我们可以发现，新来的数据距离最近的5个点，有三个点属于第一类，有两个点属于第二类，根据少数服从多数原则，新来的数据就属于第一类！

投票

from collections import CounterCounter(topk_y)

输出结果：Counter({1: 2, 0: 3})

votes = Counter(topk_y)votes.most_common(1)y_new = votes.most_common(1)[0][0]

输出结果：0

这样，我们就完成了一个基本的knn！

自己写一个knn函数

knn是一个不需要训练过程的机器学习算法。其数据集可以近似看成一个模型。

import numpy as npfrom math import sqrtfrom collections import Counter
def kNN_classifier(k, x_train, y_train, x_new):
    assert 1 <= k <= x_train.shape[0], "k must be valid"    assert x_train.shape[0] == y_train.shape[0], "the size of x_train must be equal to the size of y_train"    assert x_train.shape[1] == x_new.shape[0], "the feature number of x_new must be equal to x_train"
    distances = [sqrt(np.sum((x_new - x) ** 2)) for x in x_train]    nearest = np.argsort(distances)
    topk_y = [y_train[i] for i in nearest[:k]]    votes = Counter(topk_y)
    return votes.most_common(1)[0][0]

测试一下：

raw_data_x = [[3.3935, 2.3313],              [3.1101, 1.7815],              [1.3438, 3.3684],              [3.5823, 4.6792],              [2.2804, 2.8670],              [7.4234, 4.6965],              [5.7451


    
, 3.5340],              [9.1722, 2.5111],              [7.7928, 3.4241],              [7.9398, 0.7916]]raw_data_y = [0, 0, 0, 0, 0, 1, 1, 1, 1, 1]x_train = np.array(raw_data_x)y_train = np.array(raw_data_y)
x_new = np.array([8.0936, 3.3657])
y_new = kNN_classifier(5, x_train, y_train, x_new)print(y_new)

使用sklearn中的KNN

from sklearn.neighbors import KNeighborsClassifierimport numpy as np
raw_data_x = [[3.3935, 2.3313],              [3.1101, 1.7815],              [1.3438, 3.3684],              [3.5823, 4.6792],              [2.2804, 2.8670],              [7.4234, 4.6965],              [5.7451, 3.5340],              [9.1722, 2.5111],              [7.7928, 3.4241],              [7.9398


    
, 0.7916]]raw_data_y = [0, 0, 0, 0, 0, 1, 1, 1, 1, 1]x_train = np.array(raw_data_x)y_train = np.array(raw_data_y)
x_new = np.array([8.0936, 3.3657])
knn_classifier = KNeighborsClassifier(n_neighbors=5)knn_classifier.fit(x_train, y_train)
y_new = knn_classifier.predict(x_new.reshape(1, -1))print(y_new[0])

自己写一个面向对象的KNN

import numpy as npfrom math import sqrtfrom collections import Counter
class KNNClassifier():
    def __init__(self, k):        assert 1 <= k, "k must be valid"        self.k = k        self._x_train = None        self._y_train = None
    def fit(self, x_train, y_train):        assert x_train.shape[0] == y_train.shape[0], \         "the size of x_train must be equal to the size of y_train"        assert self.k <= x_train.shape[0], \          "the size of x_train must be at least k"


    

        self._x_train = x_train        self._y_train = y_train        return self
    def predict(self, x_new):        x_new = x_new.reshape(1, -1)        assert self._x_train is not None and self._y_train is not None, \        "must fit before predict"        assert x_new.shape[1] == self._x_train.shape[1], \        "the feature number of x must be equal to x_train"
        y_new = [self._predict(x) for x in x_new]        return np.array(y_new)
    def _predict(self, x):        assert x.shape[0] == self._x_train.shape[1], \        "the feature number of x must be equal to x_train"
        distances = [sqrt(np.sum((x_train - x) ** 2)) for x_train in self._x_train]        nearest = np.argsort(distances)
        topk_y = [self._y_train[i] for i in nearest[:self.k]]        votes = Counter(topk_y)
        return votes.most_common(1)[0][0]
    def __repr__(self):        return "KNN(k=%d)" % self.k

测试一下：

raw_data_x = [[3.3935, 2.3313],              [3.1101, 1.7815],              [1.3438, 3.3684],              [3.5823, 4.6792],              [2.2804, 2.8670],              [7.4234, 4.6965],              [5.7451, 3.5340],              [9.1722, 2.5111],              [7.7928, 3.4241],              [7.9398, 0.7916]]raw_data_y = [0, 0, 0, 0


    
, 0, 1, 1, 1, 1, 1]x_train = np.array(raw_data_x)y_train = np.array(raw_data_y)
x_new = np.array([8.0936, 3.3657])
knn_clf = KNNClassifier(6)knn_clf.fit(x_train, y_train)y_new = knn_clf.predict(x_new)print(y_new[0])

分割数据集

import numpy as npfrom sklearn import datasets

def train_test_split(x, y, test_ratio=0.2, seed=None):
    assert x.shape[0] == y.shape[0], "the size of x must be equal to the size of y"    assert 0.0 <= test_ratio <= 1.0, "test_ratio must be valid"
    if seed:        np.random.seed(seed)
    shuffle_idx = np.random.permutation(len(x))
    test_size = int(len(x) * test_ratio)    test_idx = shuffle_idx[:test_size]    train_idx = shuffle_idx[test_size:]
    x_train = x[train_idx]    y_train = y[train_idx]
    x_test = x[test_idx]    y_test = y[test_idx]
    return x_train, y_train, x_test, y_test

sklearn中鸢尾花数据测试KNN

import numpy as npfrom sklearn import datasetsfrom knn_clf import KNNClassifier
iris = datasets.load_iris()x = iris.datay = iris.target
x_train, y_train, x_test, y_test = train_test_split(x, y)my_knn_clf = KNNClassifier(k=3)my_knn_clf.fit(x_train, y_train)
y_predict = my_knn_clf.predict(x_test)print(sum(y_predict == y_test))print(sum(y_predict == y_test) / len(y_test))# 也可以使用sklearn中自带的数据集拆分方法from sklearn.model_selection import train_test_splitimport numpy as npfrom sklearn import datasetsfrom knn_clf import KNNClassifier
iris = datasets.load_iris()x = iris.datay = iris.targetx_train, y_train, x_test, y_test = train_test_split(x, y, \                                                    test_size=0.2, random_state=666)my_knn_clf = KNNClassifier(k=3)my_knn_clf.fit(x_train, y_train)y_predict = my_knn_clf.predict(x_test)print(sum(y_predict == y_test))print(sum(y_predict == y_test) / len(y_test))

sklearn中手写数字数据集测试KNN

首先，先来了解一下手写数字数据集。

import numpy as npimport matplotlib.pyplot as pltfrom sklearn import datasets
digits = datasets.load_digits()digits.keys()print(digits.DESCR)y.shapedigits.target_namesy[:100]x[:10]some_digit = x[666]y[666]some_digit_image = some_digit.reshape(8, 8)plt.imshow(some_digit_image, cmap=plt.cm.binary)plt.show()

接下来，就开始动手试试。

from sklearn import datasetsfrom shuffle_dataset import train_test_splitfrom knn_clf import KNNClassifier
digits = datasets.load_digits()x = digits.datay = digits.target
x_train, y_train, x_test, y_test = train_test_split(x, y, test_ratio=0.2)my_knn_clf = KNNClassifier(k=3)my_knn_clf.fit(x_train, y_train)y_predict = my_knn_clf.predict(x_test)
print(sum(y_predict == y_test) / len(y_test))

把求acc封装成一个函数，方便调用。

def accuracy_score(y_true, y_predict):    assert y_true.shape[0] == y_predict.shape[0], \     "the size of y_true must be equal to the size of y_predict"
    return sum(y_true == y_predict) / len(y_true)

接下来把它封装到KNNClassifier的类中。

import numpy as npfrom math import sqrtfrom collections import Counterfrom metrics import accuracy_score
class KNNClassifier():
    def __init__(self, k):        assert 1 <= k, "k must be valid"        self.k = k        self._x_train = None        self._y_train = None
    def fit(self, x_train, y_train):        assert x_train.shape[0] == y_train.shape[0], \        "the size of x_train must be equal to the size of y_train"        assert self.k <= x_train.shape[0], \        "the size of x_train must be at least k"
        self._x_train = x_train        self._y_train = y_train        return self
    def predict(self, x_new):        # x_new = x_new.reshape(1, -1)        assert self._x_train is not None and self._y_train is not None, \        "must fit before predict"        assert x_new.shape[1] == self._x_train.shape[1], \        "the feature number of x must be equal to x_train"
        y_new = [self._predict(x) for x in x_new]        return np.array(y_new)
    def _predict(self, x):        assert x.shape[0] == self._x_train.shape[1], \        "the feature number of x must be equal to x_train"
        distances = [sqrt(np.sum((x_train - x) ** 2)) for x_train in self._x_train]        nearest = np.argsort(distances)
        topk_y = [self._y_train[i] for i in nearest[:self.k]]        votes = Counter(topk_y)
        return votes.most_common(1)[0][0]
    def score(self, x_test, y_test):


    
        y_predict = self.predict(x_test)        return accuracy_score(y_test, y_predict)
    def __repr__(self):        return "KNN(k=%d)" % self.k

其实，在sklearn中这些都已经封装好了。

from sklearn.metrics import accuracy_scorefrom sklearn.model_selection import train_test_splitfrom sklearn.neighbors import KNeighborsClassifier
digits = datasets.load_digits()x = digits.datay = digits.target
x_train, x_test, y_train, y_test = train_test_split(x, y, test_size=0.2)knn_classifier = KNeighborsClassifier(n_neighbors=3)knn_classifier.fit(x_train, y_train)knn_classifier.score(x_test, y_test)

超参数

在knn中的超参数k何时最优？

from sklearn.metrics import accuracy_score from sklearn.model_selection import train_test_split from sklearn.neighbors import KNeighborsClassifier 
digits = datasets.load_digits() x = digits.data y = digits.target x_train, x_test, y_train, y_test = train_test_split(x, y, test_size=0.2) 
best_score = 0.0 best_k = -1 for k in range(1, 11):     knn_clf = KNeighborsClassifier(n_neighbors=k) 


    
    knn_clf.fit(x_train, y_train)     score = knn_clf.score(x_test, y_test)     if score > best_score:         best_k = k         best_score = score print("best k=", best_k) print("best score=", best_score)

投票方式

上面这张图，绿色的球最近的三颗球分别是红色的1号，紫色的3号和蓝色的4号。如果只考虑绿色的k个近邻中多数服从少数，目前来说就是平票。

即使不是平票，红色也是距离绿色最近。此时我们就可以考虑给他们加个权重。一般使用距离的倒数作为权重。假设距离分别为1、 3、 4

红球：1 紫+蓝：1/3 + 1/4 = 7/12

这两者加起来都没有红色的权重大，因此最终将这颗绿球归为红色类别。这样能有效解决平票问题。因此，这也算knn的一个超参数。

其实这个在sklearn封装的knn中已经考虑到了这个问题。在KNeighborsClassifier(n_neighbors=k，weights=？)

还有一个参数weights，一般有两种：uniform、distance。

from sklearn.metrics import accuracy_scorefrom sklearn.model_selection import train_test_splitfrom sklearn.neighbors import KNeighborsClassifier
digits = datasets.load_digits()x = digits.datay = digits.target
x_train, x_test, y_train, y_test = train_test_split(x, y, test_size=0.2)
best_method = ""best_score = 0.0best_k = -1for method in["uniform", "distance"]:     for k in range(1, 11):        knn_clf = KNeighborsClassifier(n_neighbors=k, weights=method)        knn_clf.fit(x_train, y_train)        score = knn_clf.score(x_test, y_test)        if score > best_score:            best_method = method            best_k = k            best_score = scoreprint("best_method=", best_method)print("best k=", best_k)print("best score=", best_score)

如果使用距离，那么有很多种距离可以使用，欧氏距离、曼哈顿距离、明可夫斯基距离。

from sklearn.metrics import accuracy_scorefrom sklearn.model_selection import train_test_splitfrom sklearn.neighbors import KNeighborsClassifier
digits = datasets.load_digits()x = digits.datay = digits.target
x_train, x_test, y_train, y_test = train_test_split(x, y, test_size=0.2)
best_p = -1best_score = 0.0best_k = -1for p in range(1, 6):    for k in range(1, 11):        knn_clf = KNeighborsClassifier(n_neighbors=k, weights="distance", p=p)        knn_clf.fit(x_train, y_train)        score = knn_clf.score(x_test, y_test)        if score > best_score:            best_p = p            best_k = k            best_score = scoreprint("best_p=", best_p)print("best k=", best_k)print("best score=", best_score)

下载1：OpenCV-Contrib扩展模块中文版教程

在「小白学视觉」公众号后台回复：扩展模块中文教程，即可下载全网第一份OpenCV扩展模块教程中文版，涵盖扩展模块安装、SFM算法、立体视觉、目标跟踪、生物视觉、超分辨率处理等二十多章内容。

下载2：Python视觉实战项目52讲

在「小白学视觉」公众号后台回复：Python视觉实战项目，即可下载包括图像分割、口罩检测、车道线检测、车辆计数、添加眼线、车牌识别、字符识别、情绪检测、文本内容提取、面部识别等31个视觉实战项目，助力快速学校计算机视觉。

下载3：OpenCV实战项目20讲

在「小白学视觉」公众号后台回复：OpenCV实战项目20讲，即可下载含有20个基于OpenCV实现20个实战项目，实现OpenCV学习进阶。

交流群

欢迎加入公众号读者群一起和同行交流，目前有SLAM、三维视觉、传感器、自动驾驶、计算摄影、检测、分割、识别、医学影像、GAN、算法竞赛等微信群（以后会逐渐细分），请扫描下面微信号加群，备注：”昵称+学校/公司+研究方向“，例如：”张三 + 上海交大 + 视觉SLAM“。请按照格式备注，否则不予通过。添加成功后会根据研究方向邀请进入相关微信群。请勿在群内发送广告，否则会请出群，谢谢理解~