機器學習實戰之KMeans算法pandas實現

這次寫的噁心死我了，第一次隨機選中心點的代碼有問題還是怎麼的，導致第一輪過完，可能會出現某個中心點根本就完全不合適，沒有任何一個點會標記爲這個中心點，然後報錯。導致我的這個代碼時靈時不靈，開始完全想不到bug的原因。
而且雖然用pandas來自己實現確實能幫忙鞏固pandas的知識，但我還是覺得以前是走入了誤區。機器學習重要的部分應該是對理論的理解和輪子的使用，至於書中的代碼，理解下理論實現的具體過程就行了，自己在這費心費力造輪子實在是意義不大，雖然在造的過程中也能加深理解，但事倍功半
所以還是決定以後學習就結合西瓜書的理論，實戰中的實現思路和sklearn的代碼實現了

import numpy as np
import pandas as pd
from pandas import DataFrame,Series

path = r'C:\Users\36955\Downloads\mlp\Ch10\testSet2.txt'
data = pd.read_csv(path,sep='\t',header=None)

def randCent(dataSet, k):
    n = dataSet.shape[1]
    centroids = DataFrame(np.zeros((k,n)))
    data_min = dataSet.min(0)
    data_range = dataSet.max(0)-data_min
    for j in range(n):
        minJ = data_min[j]
        rangeJ = float(data_range[j])
        centroids.iloc[:,j] = minJ + rangeJ * np.random.rand(k,1)
    return centroids

def distEclud(vecA, vecB):
    return np.sqrt(np.sum(np.power(vecA - vecB, 2)))

def kMeans(data,k):
    m = data.shape[0]
    clusterAssment = DataFrame(np.zeros((m,2)),columns=['clusterName','dist'])
    centroids = randCent(data,k)
    clusterChanged = True
    while clusterChanged:
        clusterChanged = False
        for i in range(m):
            minDist = np.inf;minIndex = -1
            for j in range(k):
                #print (i,j,m,len(centroids))
                dist = distEclud(data.iloc[i,:],centroids.iloc[j,:])
                if dist < minDist:
                    minDist = dist;minIndex = j
            if clusterAssment.iloc[i,0] != minIndex:
                clusterChanged = True
            clusterAssment.iloc[i,:] = minIndex,minDist**2
        #print centroids
        centroids = data.groupby(clusterAssment.clusterName).mean()
    return centroids,clusterAssment

#print kMeans(data,3)

def biKmeans(data,k):
    m = data.shape[0]
    clusterAssment = DataFrame(np.zeros((m,2)),columns=['clusterName','dist'])
    centroid0 = list(data.mean(0))
    centList = [centroid0]
    while len(centList)<k:
        lowestSSE = np.inf
        for i in range(len(centList)):
            splitCluster = data.iloc[clusterAssment[clusterAssment.clusterName==i].index,:]
            splitCentroids,splitClusterAssement = kMeans(splitCluster,2)
            splitSSE = splitClusterAssement.dist.sum()
            nosplitSSE = clusterAssment[clusterAssment.clusterName!=i].dist.sum()
            if (splitSSE+nosplitSSE)<lowestSSE:
                besttosplit = i
                newCentroids = splitCentroids.copy()
                newAssement = splitClusterAssement.copy()
                lowestSSE = splitSSE+nosplitSSE
        centList[besttosplit] = list(newCentroids.iloc[0,:])
        centList.append(list(newCentroids.iloc[1,:]))
        newAssement.clusterName[newAssement.clusterName==0] = besttosplit
        newAssement.clusterName[newAssement.clusterName==1] = len(centList)-1
        clusterAssment=clusterAssment[clusterAssment.clusterName!=besttosplit]
        clusterAssment = pd.concat([clusterAssment,newAssement],axis=0)
    return centList,clusterAssment

print biKmeans(data,4)