机器学习技法笔记：Homework #8 kNN&RBF&k-Means相关习题

原文地址：https://www.jianshu.com/p/1db700f866ee

问题描述

程序实现

# kNN_RBFN.py
# coding:utf-8

import numpy as np
import matplotlib.pyplot as plt

def ReadData(dataFile):

    with open(dataFile, 'r') as f:
        lines = f.readlines()
        data_list = []
        for line in lines:
            line = line.strip().split()
            data_list.append([float(l) for l in line])
        dataArray = np.array(data_list)
        return dataArray

def sign(n):

    if(n>=0):
        return 1
    else:
        return -1

def kNN(k,trainArray,dataX):
    num_data=dataX.shape[0]
    predY=np.zeros((num_data,))
    for n in range(num_data):
        distArray=np.sum((trainArray[:,:-1]-dataX[n,:])**2,axis=1)
        id_list=np.argsort(distArray,axis=0).tolist()[:k]
        for i in id_list:
            predY[n]+=trainArray[i,-1]
        predY[n]=sign(predY[n])
    return predY

def GetZeroOneError(predY,dataY):
    return (predY!=dataY).sum()/dataY.shape[0]

def plot_bar_chart(X,Y,nameX,nameY,saveName):
    plt.figure(figsize=(10,6))
    plt.bar(left=X,height=Y,width=0.8,align="center",yerr=0.000001)
    for (c,w) in zip(X,Y):
        plt.text(c,w*1.03,str(round(w,4)))
    plt.xlabel(nameX)
    plt.ylabel(nameY)
    plt.xlim(X[0]-1,X[-1]+1)
    plt.xticks(X)
    plt.ylim(0,1)
    plt.title(nameY+" versus "+nameX)
    plt.savefig(saveName)
    return

def RBFNetwork(k,gamma,trainArray,dataX):
    num_data=dataX.shape[0]
    predY=np.zeros((num_data,))
    for n in range(num_data):
        gaussianDistArray=np.exp(-gamma*np.sum((trainArray[:,:-1]-dataX[n,:])**2,axis=1))
        id_list=np.argsort(gaussianDistArray,axis=0).tolist()[:k]
        for i in id_list:
            predY[n]+=trainArray[i,-1]
        predY[n]=sign(predY[n])
    return predY

if __name__=="__main__":

    dataArray=ReadData("hw8_train.dat")
    testArray=ReadData("hw8_test.dat")
    k_list=[1,3,5,7,9]
    ein_list=[]
    eout_list=[]
    for k in k_list:
        predY=kNN(k,dataArray,dataArray[:,:-1])
        ein_list.append(GetZeroOneError(predY,dataArray[:,-1]))
        predY=kNN(k,dataArray,testArray[:,:-1])
        eout_list.append(GetZeroOneError(predY,testArray[:,-1]))

    # 12
    plot_bar_chart(k_list,ein_list,nameX="k",nameY="Ein(gk-nbor)",saveName="12.png")

    # 14
    plot_bar_chart(k_list,eout_list,nameX='k',nameY="Eout(gk-bor)",saveName="14.png")

    gamma_list=[-3,-1,0,1,2]
    ein_list=[]
    eout_list=[]
    for gamma in gamma_list:
        predY=RBFNetwork(dataArray.shape[0],10**gamma,dataArray,dataArray[:,:-1])
        ein_list.append(GetZeroOneError(predY,dataArray[:,-1]))
        predY=RBFNetwork(dataArray.shape[0],10**gamma,dataArray,testArray[:,:-1])
        eout_list.append(GetZeroOneError(predY,testArray[:,-1]))

    # 16
    plot_bar_chart(X=gamma_list,Y=ein_list,nameX="log10(gamma)",nameY="Ein(guniform)",saveName="16.png")

    # 18
    plot_bar_chart(X=gamma_list,Y=eout_list,nameX="log10(gamma)",nameY="Eout(guniform)",saveName="18.png")

# kMeans.py
# coding:utf-8

from numpy import random
from kNN_RBFN import *

def kMeans(t,k,dataArray):
    num_data=dataArray.shape[0]
    random.seed(t)
    centreIDList=random.randint(0,num_data,k).tolist()
    nowCentreArray=dataArray[centreIDList,:]
    tmpCentreArray=np.array(nowCentreArray)
    ein=1000000
    nowEin=ein-1
    dict={}
    while(nowEin<ein):
        ein=nowEin
        dict = {}
        for n in range(num_data):
            distArray=np.sum((nowCentreArray-dataArray[n,:])**2,axis=1)
            minID=np.argmin(distArray)
            tmpCentreArray[minID]=(tmpCentreArray[minID]+dataArray[n,:])/2
            try:
                dict[minID].append(dataArray[n,:])
            except:
                dict[minID]=[]
                dict[minID].append(dataArray[n,:])
        nowCentreArray=np.array(tmpCentreArray)
        nowEin=GetEin(nowCentreArray,dict)
    return nowCentreArray,dict

def GetEin(nowCentreArray,dict):
    k=nowCentreArray.shape[0]
    ein=0
    for i in range(k):
        if i not in dict.keys():
            continue
        data=np.array(dict[i])
        ein+=np.average(np.sum((data-nowCentreArray[i])**2,axis=1))
    return ein

def plot_bar_chart(X,Y,nameX,nameY,saveName):
    plt.figure(figsize=(10,6))
    plt.bar(left=X,height=Y,width=0.8,align="center",yerr=0.000001)
    for (c,w) in zip(X,Y):
        plt.text(c,w*1.03,str(round(w,4)))
    plt.xlabel(nameX)
    plt.ylabel(nameY)
    plt.xlim(X[0]-1,X[-1]+1)
    plt.xticks(X)
    plt.title(nameY+" versus "+nameX)
    plt.savefig(saveName)
    return

if __name__=="__main__":

    dataArray=ReadData("hw8_nolabel_train.dat")
    k_list=[2,4,6,8,10]
    ein_list=[]
    for k in k_list:
        ein=0
        for t in range(500):
            nowCentreArray,dict=kMeans(t,k,dataArray)
            ein+=GetEin(nowCentreArray,dict)
        ein_list.append(ein/500)

    plot_bar_chart(k_list,ein_list,nameX="k",nameY="the average Ein over 500 experiments",saveName="20.png")

运行结果