机器学习实战02——kNN（2）

说明

《机器学习实战》第20-28页的代码,含部分注释

代码

# Author : hang
# @TIme  : 2021-11-25 20:31
# @File  : kNN2.py

from numpy import *
import matplotlib.pyplot as plt

def classify(inX,dataSet,labels,k):
    rows = dataSet.shape[0]
    inX = tile(inX,(rows,1))
    inX = inX-dataSet
    inX = (inX**2).sum(axis=1)
    distance = inX**0.5
    index = distance.argsort()
    dict = {}
    for i in range(k):
        pos = labels[index[i]]
        dict[pos] = dict.get(pos,0)+1
    Dict = sorted(dict.items(),key=lambda x:x[1],reverse=True)
    return Dict[0][0]

def filetoInfo(filename): # 从目标文件中提取数据
    fr = open(filename)
    lines = len(fr.readlines())
    mat = zeros((lines,3))
    labels = []
    index = 0
    fr = open(filename)
    for line in fr.readlines():
        line = line.strip().split('\t')
        mat[index,:] = line[0:3]
        labels.append(int(line[-1]))
        index += 1
    return array(mat),array(labels)


def autoNorm(dataSet): # 归一化
    minVals = dataSet.min(0) # 求每列的最小值
    maxVals = dataSet.max(0) # 求每列的最大值
    ranges = maxVals - minVals
    m = dataSet.shape[0]  # dataSet的行数
    up = (dataSet - tile(minVals,(m,1))) # 分子
    down = tile(ranges,(m,1)) # 分母
    normDataSet = up/down
    return normDataSet,ranges,minVals

def datingClassTest():
    mat,labels = filetoInfo('datingTestSet2.txt')
    m = mat.shape[0]
    testNum = int(m*0.1)
    errcnt = 0
    for i in range(testNum):
        result = classify(mat[i,:],mat[testNum:m,:],labels[testNum:m],10)
        if result != labels[i]:
            errcnt += 1
    print("error cnt: %d  error rate: %f" % (errcnt,errcnt/float(testNum)))

def datingClassTest():
    testRate = 0.5
    mat,labels = filetoInfo('datingTestSet2.txt')
    normMat,ranges,minVals = autoNorm(mat)
    m = normMat.shape[0]
    test = int(m*testRate)
    errCnt = 0
    for i in range(test):
        result = classify(normMat[i,:],normMat[test:m,:],labels[test:m],3)
        print("result: %d  real: %d" % (result,labels[i]))
        if result!=labels[i]:
            errCnt += 1
    print("err rate: %f" % (errCnt/float(test)))
    print("err count: %d" % (errCnt))

# Mat, labels = filetoInfo('datingTestSet2.txt')
# print(Mat)
# print(labels)

# fig = plt.figure()
# ax = fig.add_subplot(111)
# ax.scatter(Mat[:,1],Mat[:,2],15*labels,labels)
# plt.show()

# normMat,ranges,minVals = autoNorm(Mat)
# print(normMat)
# print(ranges)
# print(minVals)

datingClassTest()

注意

dataSet的数据格式要保持一致

总结

kNN算法逻辑较为简单，大家可以利用debug模式逐步查看每个变量，就能逐渐理解代码前后的逻辑