decisionTrees_ID3

import math,operator

def calShannonEnt(dataSet):
    numEntries = len(dataSet)
    labelCounts = {}
    for featVec in dataSet:
        currentLabel = featVec[-1]
        if currentLabel not in labelCounts.keys():
            labelCounts[currentLabel] = 0
        labelCounts[currentLabel] +=1
    shannonEnt = 0.0
    for key in labelCounts:
        prob = float(labelCounts[key])/numEntries
        shannonEnt -= prob*math.log(prob,2)
    return shannonEnt

def createDataSet():
    dataSet = [[1,1,'yes'],[1,1,'yes'],[1,0,'no'],[0,1,'no'],[0,1,'no']]
    labels = ['no surfacing','flippers']
    return dataSet,labels

def splitDataSet(dataSet,axis,value):
    retDataSet = []
    for featVec in dataSet:
        if featVec[axis] == value:
            reducedFeatVec = featVec[:axis]
            reducedFeatVec.extend(featVec[axis+1:])
            retDataSet.append(reducedFeatVec)
    return retDataSet

def chooseBestFeatureToSplit(dataSet):
    numFeatures = len(dataSet[0])-1
    baseEntropy = calShannonEnt(dataSet)
    bestInfoGain = 0.0; bestFeature = -1
    featList = []
    for i in range(numFeatures):
        featList = [example[i] for example in dataSet]
        uniqueVals = set(featList)
        newEntropy = 0.0
        for value in uniqueVals:
            subDataSet = splitDataSet(dataSet,i,value)
            pro = len(subDataSet)/len(dataSet)
            newEntropy += pro*calShannonEnt(subDataSet)
        infoGain = baseEntropy - newEntropy
        if(infoGain>bestInfoGain):
            bestInfoGain = infoGain
            bestFeature = i
    return bestFeature

def majorityCnt(classList):
    classCout = {}
    for vote in classList:
        if vote not in classCout.keys():classCout[vote] =0
        classCout[vote]+=1
    sortedClassCout = sorted(classCout.iteritems(),key=operator.itemgetter(1),reverse=True)
    return sortedClassCout[0][0]


def createTree(dataSet,labels):
    classList = [example[-1] for example in dataSet]
    if classList.count(classList[0]) == len(classList):
        return classList[0]#all the class label are the same
    if len(dataSet[0]) == 1:#calculate the feature number
        return majorityCnt(classList)#no more class features to split
    bestFeat = chooseBestFeatureToSplit(dataSet)
    bestFeatLabel = labels[bestFeat]
    myTree = {bestFeatLabel:{}}
    del(labels[bestFeat])
    featValues = [example[bestFeat] for example in dataSet]
    uniqueVals = set(featValues)
    for value in uniqueVals:
        subLabel = labels[:]
        myTree[bestFeatLabel][value]=createTree(splitDataSet(dataSet,bestFeat,value),subLabel)
    return myTree

def classfy(inputTree,featLabels,testVec):
    firstStr = inputTree.keys()[0]
    secondDict = inputTree[firstStr]
    featIndex = featLabels.index(firstStr)
    for key in secondDict.keys():
        if testVec[featIndex] == key:
            if type(secondDict[key]).__name__ =='dict':
                classLabel = classfy(secondDict[key],featLabels,testVec)
            else: classLabel = secondDict[key]
    return classLabel

def storeTree(inputTree,filename):
    import pickle
    fw = open(filename,'w')
    pickle.dump(inputTree,fw)
    fw.close()

def grabTree(filename):
    import pickle
    fr = open(filename)
    return pickle.load(fr)

#fr=open('.txt')
#lines = fr.readlines()
#dataset = [line.strip().split('\t') for line in lines]
#labels = [line[-1] for line in dataSet]
#lenseLabels = ['age','prescript','astigmatic','tearRate']