- 最近邻法和k-近邻法
下面图片中只有三种豆，有三个豆是未知的种类，如何判定他们的种类？
提供一种思路，即：未知的豆离哪种豆最近就认为未知豆和该豆是同一种类。由此，我们引出最近邻算法的定义：为了判定未知样本的类别，以全部训练样本作为代表点，计算未知样本与所有训练样本的距离，并以最近邻者的类别作为决策未知样本类别的唯一依据。但是，最近邻算法明显是存在缺陷的，比如下面的例子：有一个未知形状(图中绿色的圆点)，如何判断它是什么形状？
显然，最近邻算法的缺陷——对噪声数据过于敏感，为了解决这个问题，我们可以可以把未知样本周边的多个最近样本计算在内，扩大参与决策的样本量，以避免个别数据直接决定决策结果。由此，我们引进K-最近邻算法。K-最近邻算法是最近邻算法的一个延伸。基本思路是：选择未知样本一定范围内确定个数的K个样本，该K个样本大多数属于某一类型，则未知样本判定为该类型。如何选择一个最佳的K值取决于数据。一般情况下，在分类时较大的K值能够减小噪声的影响，但会使类别之间的界限变得模糊。待测样本（绿色圆圈）既可能分到红色三角形类，也可能分到蓝色正方形类。如果k取3，从图可见，待测样本的3个邻居在实线的内圆里，按多数投票结果，它属于红色三角形类。但是如果k取5，那么待测样本的最邻近的5个样本在虚线的圆里，按表决法，它又属于蓝色正方形类。在实际应用中，K先取一个比较小的数值，再采用交叉验证法来逐步调整K值，最终选择适合该样本的最优的K值。
- KNN算法实现
算法基本步骤：
1）计算待分类点与已知类别的点之间的距离
2）按照距离递增次序排序
3）选取与待分类点距离最小的k个点
4）确定前k个点所在类别的出现次数
5）返回前k个点出现次数最高的类别作为待分类点的预测分类
下面是一个按照算法基本步骤用python实现的简单例子，根据已分类的4个样本点来预测未知点(图中的灰点)的分类：
from numpy import * # create a dataset which contains 4 samples with 2 classes def createDataSet(): # create a matrix: each row as a sample group = array([[1.0, 0.9], [1.0, 1.0], [0.1, 0.2], [0.0, 0.1]]) labels = ['A', 'A', 'B', 'B'] # four samples and two classes return group, labels # classify using kNN (k Nearest Neighbors ) # Input: newInput: 1 x N # dataSet: M x N (M samples N, features) # labels: 1 x M # k: number of neighbors to use for comparison # Output: the most popular class label def kNNClassify(newInput, dataSet, labels, k): numSamples = dataSet.shape[0] # shape[0] stands for the num of row ## step 1: calculate Euclidean distance # tile(A, reps): Construct an array by repeating A reps times # the following copy numSamples rows for dataSet diff = tile(newInput, (numSamples, 1)) - dataSet # Subtract element-wise squaredDiff = diff ** 2 # squared for the subtract squaredDist = sum(squaredDiff, axis = 1) # sum is performed by row distance = squaredDist ** 0.5 ## step 2: sort the distance # argsort() returns the indices that would sort an array in a ascending order sortedDistIndices = argsort(distance) classCount = {} # define a dictionary (can be append element) for i in xrange(k): ## step 3: choose the min k distance voteLabel = labels[sortedDistIndices[i]] ## step 4: count the times labels occur # when the key voteLabel is not in dictionary classCount, get() # will return 0 classCount[voteLabel] = classCount.get(voteLabel, 0) + 1 ## step 5: the max voted class will return maxCount = 0 for key, value in classCount.items(): if value > maxCount: maxCount = value maxIndex = key return maxIndex if __name__== "__main__": dataSet, labels = createDataSet() testX = array([1.2, 1.0]) k = 3 outputLabel = kNNClassify(testX, dataSet, labels, 3) print "Your input is:", testX, "and classified to class: ", outputLabel testX = array([0.1, 0.3]) outputLabel = kNNClassify(testX, dataSet, labels, 3) print "Your input is:", testX, "and classified to class: ", outputLabel
结果如下：
Your input is: [ 1.2 1. ] and classified to class: A
Your input is: [ 0.1 0.3] and classified to class: B
OpenCV中也提供了机器学习的相关算法，其中KNN算法的最基本例子如下
import numpy as np import matplotlib.pyplot as plt import cv2 # Feature set containing (x,y) values of 25 known/training data trainData = np.random.randint(0,100,(25,2)).astype(np.float32) # Labels each one either Red or Blue with numbers 0 and 1 responses = np.random.randint(0,2,(25,1)).astype(np.float32) # Take Red families and plot them red = trainData[responses.ravel()==0] plt.scatter(red[:,0],red[:,1],80,'r','^') # Take Blue families and plot them blue = trainData[responses.ravel()==1] plt.scatter(blue[:,0],blue[:,1],80,'b','s') # Testing data newcomer = np.random.randint(0,100,(1,2)).astype(np.float32) plt.scatter(newcomer[:,0],newcomer[:,1],80,'g','o') knn = cv2.KNearest() knn.train(trainData,responses) # Trains the model # Finds the neighbors and predicts responses for input vectors. ret, results, neighbours ,dist = knn.find_nearest(newcomer, 3) print "result: ", results,"\n" print "neighbours: ", neighbours,"\n" print "distance: ", dist plt.show()