KNN — K近邻算法

Demo

# -*- coding: utf-8 -*-
'''
Project: KNNDemo
File: demo.py
Author: Weifu Liu
Time:2020.06.12
'''
import numpy as np
import operator
import matplotlib.pyplot as plt

def createDataSet():
    '''
    Input:None
    Output:Group--a tran-dataset
           ClassLabelVector--the label of class used to plot scatter color
    Function:Prepare data       
    '''
    group = np.array([[1.0,1.1],[1.0,1.0],[0,0],[0,0.1]])
    labels = ['A','A','B','B']
    classLabelVector = []
    for label in labels:
        if label == 'A':
            classLabelVector.append(1)
        if label == 'B':
            classLabelVector.append(2)
    return group,classLabelVector

def createScatter(dataset,classLabelVector):
    '''
    Input: Sample Dataset
    Output: A Scatter Plot
    Function:Plot a scatter plot from sample dataset
    '''
    print(dataset)
    plt.scatter(dataset[:,0], dataset[:,1],alpha=0.6,c=classLabelVector)
    plt.ylabel('X')
    plt.xlabel('y')
    plt.title("KNN Demo")
    plt.show()

def classify0(inX,dataSet,labels,k):
    '''
    Input: inX - test vector
           dataset -test-dataset
           labels - categories
           k - parameter K of KNN algorithm 
    Output:result of KNN
    '''
    dataSetSize = dataSet.shape[0]
    print(dataSetSize)
    # diffMat is the value of x[i]-x[j] and y[i]-y[j]
    diffMat = np.tile(inX,(dataSetSize,1)) - dataSet
    print(diffMat)
    sqDiffMat = diffMat**2
    sqDistances = sqDiffMat.sum(axis=1)
    print(sqDistances)
    distances = sqDistances**0.5
    sortedDistIndicies = distances.argsort()
    classCount = {}
    for i in range(k):
        voteILabel = labels[sortedDistIndicies[i]]
        classCount[voteILabel] = classCount.get(voteILabel,0) + 1
    sortedClassCount = sorted(classCount.items(),key=operator.itemgetter(1),reverse=True)
    return sortedClassCount[0][0]        
    
    
def main():
    group,classLabelVector = createDataSet()
    print(group[0])
    createScatter(group,classLabelVector)
    testVec = np.array([1.2,1.1])
    k = 3
    print("测试数据的类别为：",classify0(testVec, group, classLabelVector, k))
    
if __name__ == "__main__":
    main()
    

鸢尾花测试

import numpy as np
import operator
import matplotlib.pyplot as plt
import csv
import random


def PreProcess(path):
    '''
    Input: path - the file path of dataset
    Output:None
    Function: get the tran_data and test_data
    '''
    f = open(path,'r',encoding='utf8')
    tran_csv = open('data/tran_data.csv','w',encoding='utf8',newline='')
    test_csv = open('data/test_data.csv','w',encoding='utf8',newline='')
    tran_writer = csv.writer(tran_csv)
    test_writer = csv.writer(test_csv)
    reader = csv.reader(f)
    for i in reader:
        print(i)
        rand = random.randint(1,5)
        lst = []
        lst.append(i[0])
        lst.append(i[1])
        lst.append(i[2])
        lst.append(i[3])
        if i[4] == 'Iris-setosa':
            lst.append('1')
        if i[4] == 'Iris-versicolor':
            lst.append('2')
        if i[4] == 'Iris-virginica':
            lst.append('3')
        if rand == 1:
            test_writer.writerow(lst)
        else:
            tran_writer.writerow(lst)
        
        
def LoadDataset():
    '''
    Input:None
    Output:tran_data,test_data
    Function:Load tran_data and test_data
    '''
    tran_data = np.loadtxt('data/tran_data.csv',delimiter=',')
    test_data = np.loadtxt('data/test_data.csv',delimiter=',')
    #print(dataset)
    return tran_data,test_data

def calDistance(inX,dataset,k):
    '''
    Input:Inx - test vector
          dataset - tran_data
          k - KNN K value
    Output:Predicted class
    Function:Calculate the distance to get the distance matrix
    '''
    repeatTime = dataset.shape[0]
    diffMat = np.tile(inX,(repeatTime,1)) - dataset
    #print(diffMat)
    sqDiffMat = diffMat**2
    sqDistances = sqDiffMat.sum(axis=1) - sqDiffMat[:,4]
    distances = sqDistances**0.5
    #print(distances)
    return KNNcls(distances,k,dataset)
        
def KNNcls(distances,k,dataset):
    '''
    Input: distances - the distance matrix
           k - KNN K value
           dataset - tran_data
    Output:Predicted class     
    Function:perform KNN classification to get the prediction result
    '''
    sortedDistanceIndex = distances.argsort()
    #print(sortedDistanceIndex)
    voteCount = {}
    Labels ={"1":"Iris-setosa","2":"Iris-versicolor","3":"Iris-virginica"}
    for i in range(k):
        labelID = dataset[sortedDistanceIndex[i]][4]
        #print("class ID is:",int(labelID))
        labelName = Labels[str(int(labelID))]
        #print("class label is:",labelName)
        voteCount[labelName] = voteCount.get(labelName,0) + 1
    #print(voteCount)
    sortedVoteCount = sorted(voteCount.items(),key=operator.itemgetter(1),reverse=True)
    #print('predicted = ',sortedVoteCount[0][0])
    return sortedVoteCount[0][0]
    
def main():
    ''' 
    Input:None
    Output:None
    Function:Use the classifier to test the data set for accuracy
    '''
    #path = 'data/iris.csv'
    #PreProcess(path)
    tran_data,test_data = LoadDataset()
    k = 3
    Labels ={"1":"Iris-setosa","2":"Iris-versicolor","3":"Iris-virginica"}
    sumCount = 0
    correctCount = 0
    for inX in test_data:
        print(inX)
        actualLabel = Labels[str(int(inX[4]))]
        print('actual = ',actualLabel)
        predictedLabel = calDistance(inX,tran_data,k)
        print('predicted = ',predictedLabel)
        sumCount += 1
        if actualLabel == predictedLabel:
            correctCount += 1
    print('Accuracy:',correctCount/sumCount)
    #print(test_data)
    inY = [4.9, 3.0 , 1.4, 0.2,2.0]
    preLabel = calDistance(inY,tran_data,k)
    print('predicted = ',preLabel)
    
    
if __name__ == "__main__":
    main()

测试数据

训练数据

result