理想環(huán)境下的kNN

計(jì)算距離 ->取最近的K個(gè)值 -> 選取占比高的類別

1. 計(jì)算距離

原始數(shù)據(jù):9條數(shù)據(jù),2個(gè)屬性,2個(gè)類別

對于待分類的點(diǎn)**(61,20) **, 可以得到對于每個(gè)點(diǎn)的距離,使用曼哈頓距離公式.我不是為了偷懶,就是用這個(gè)

2. 最近k個(gè)

最近的三個(gè)依次是X5-B,X3-A,X1-A

3. 判斷類別

A占66%,B占33%
所以待分類點(diǎn)**(61,20) **是A類別

4. 簡單py實(shí)現(xiàn)

計(jì)算機(jī)負(fù)責(zé)計(jì)算,可以舍棄曼哈頓距離了,用最經(jīng)典的歐幾里得距離

# -*- coding: utf-8 -*-

__author__ = 'Matter-YYF'

from numpy import *
import operator

def createDataSet():
    # numpy的array可以初始化矩陣
    group = array([[50.0,20.0],[53.0,32.0],[60.0,25.0],
                  [65.0,30.0],[58.0,20.0],[10.0,50.0],
                  [20.0,47.0],[15.0,60.0],[25.0,55.0]])
    label = ['A','A','A','A','B','B','B','B','B',]
    return group,label

def classify0(inX, dataSet, labels, k):
    # shape屬性是行列數(shù)
    dataSetSize = dataSet.shape[0]
    # 用tile創(chuàng)建重復(fù)數(shù)組,此處用于創(chuàng)建矩陣 tile([61.0,20.0],(9,1))
    diffMat = tile(inX,(dataSetSize,1))-dataSet

    sqDiffMat = diffMat ** 2
    # 矩陣每行相加
    sqDistance = sqDiffMat.sum(axis=1)
    distance = sqDistance ** 0.5
    # 字典key排序
    sortedDistIndex = distance.argsort()
    classCount = {}
    for i in range(k):
        voteLabel = labels[sortedDistIndex[i]]
        # XX.get(a,b):有key=a的則取對應(yīng)value,沒有則為b
        classCount[voteLabel] = classCount.get(voteLabel,0)+1
    # 前k個(gè)類別的排序
    sortedClassCount = sorted(classCount.iteritems(),
                              key=operator.itemgetter(1),
                              reverse=True)
    # 返回占比例最高的類別
    return sortedClassCount[0][0]

dataSet,label = createDataSet()
resultKNN = classify0([61.0,20.0],dataSet,label,3)
print resultKNN

歡迎回到現(xiàn)實(shí)世界

1. 手輸數(shù)據(jù)是有多蛋疼...

勞資是程序猿,不干low活

得從文件讀數(shù)據(jù)啊

def file2matrix(filename):
    fr = open(filename)
    arrayOfLines = fr.readlines()
    numberOfLines = len(arrayOfLines)
    returnMat = zeros((numberOfLines,3))    # 生成一個(gè)空白矩陣,3列
    classLabelVector = []
    index = 0
    for line in arrayOfLines:
        line = line.strip()                 # 刪除空白符:'\n','\r','\t',' '
        listFromLine = line.split('\t')     # 以制表符為分割生成列表
        returnMat[index,:] = listFromLine[0:3]              # 前三個(gè)項(xiàng)目是數(shù)據(jù)屬性
        classLabelVector.append(int(listFromLine[-1]))      # 最后是類別標(biāo)簽
        index+=1
    return returnMat, classLabelVector

2. 數(shù)據(jù)怎么跨度這么大...

10w和25在一起,好別扭

歸歸歸歸一化

線性函數(shù)歸一化(Min-Max scaling)

def autoNorm(dataSet):
    minVals = dataSet.min(0)        # 所有列里各自最小值
    maxVals = dataSet.max(0)        # 所有列里各自最大值

    ranges = maxVals -minVals       # 歸一化計(jì)算公式分母
    normDataSet = zeros(shape(dataSet))

    m = dataSet.shape[0]            # 數(shù)據(jù)行數(shù)
    normDataSet = dataSet-tile(minVals,(m,1))       # 分子
    normDataSet = normDataSet/tile(ranges,(m,1))    #計(jì)算結(jié)果
    return normDataSet, ranges, minVals