Learning Log---Decision Tree algorithm ID3

ID3 algorithm

#coding =utf-8from math import log  import operator    # This defines a sample set Def createdataset ():      dataset = [[1, 1,  ' Yes '],                 [1 , 1,  ' yes '],                  [1, 0,  ' No '],                  [0, 1,  ' No '],                  [0, 1,  ' No ']]       labels = [' no surfacing ', ' flippers ']       #change  to  discrete values      return dataSet, labels      #这里计算该样本的期望值def  calCshannonent (DataSet):       numentries = len (DataSet)        labelCounts = {}      for featVec in  dataset:  #the  the number of unique elements and their occurance           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 * log (prob,2)   #log  base 2       return shannonEnt        #这里计算数据集中某列满足相同条件下的其他数值组成的矩阵, The expected value Def splitdataset (dataset, axis, value) for calculating the eigenvalue:       retdataset = []      for featvec in dataset:           if featVec[axis] == value:               reducedfeatvec = featvec[: axis]      #chop  out axis used for splitting               reducedfeatvec.extend (featVec[axis+1:] )               retdataset.append ( Reducedfeatvec)       return retdataset         #针对上个函数算出的矩阵, calculates the expected value of all eigenvalues, and then draws the maximum information increment def  choosebestfeaturetosplit (DataSet):       numfeatures = len ( Dataset[0])  - 1       #the  last column is used  for the labels      baseentropy = calcshannonent ( DataSet)       bestinfogain = 0.0; bestfeature = -1       for i in range (numfeatures):          #iterate  over all the features           featlist = [example[i] for example in dataset] #create  a  list of all the examples of this feature           uniqueVals = set (featlist)         #get  a set of  unique values          newentropy = 0.0           for value in uniqueVals:               #这里选取某一特征值得所有可能计算期望值               subdataset = splitdataset (DataSet, i,  value)               prob =  len (Subdataset)/float (len (dataSet))                newentropy += prob * calcshannonent (Subdataset)                 infogain = baseentropy -  newentropy      #calculate  the info gain; ie reduction in entropy           if  (Infogain > bestinfogain):         #compare  this to the best gain so far               bestInfoGain =  infogain          #if  better than current best , set to best               bestFeature = i      return bestFeature                         #returns  an integer       #如果最后的结果有多个, voting mechanism to take the biggest def  MAJORITYCNT (classlist):    &Nbsp;  classcount={}      for vote in classlist:           if vote not in classcount.keys ():  classcount[vote] = 0          classcount[ Vote] += 1      sortedclasscount = sorted ( Classcount.iteritems (),  key=operator.itemgetter (1),  reverse=true)        return sortedClassCount[0][0]     #创建树, using recursive Def createtree (dataset,labels):        #计算dataset里是否只有单值     classlist = [example[-1 ] for example in dataset]          #如果只有单值, and the only , a single value of     if classlist.count (Classlist[0])  == len (classList):   is returned          return classlist[0]            # If it is the final result, but has multiple values, take the most     if len (Dataset[0])  == 1:          RETURN MAJORITYCNT (classlist)                 #取最佳的特征值     bestFeat =  Choosebestfeaturetosplit (DataSet)       bestfeatlabel = labels[bestfeat ]          #根据这个特征值制定字典      mytree =  {bestFeatLabel:{}}           #删除这个特征值      del (Labels[bestfeat])            #找出这个特征值下有几个选择      featValues = [example[bestFeat] for example in dataSet]       uniquevals&nBsp;= set (featvalues)       for value in uniqueVals:           subLabels = labels[:]            #针对每个选择, establishing different branches               mytree[bestfeatlabel][value] = createtree (SplitDataSet (dataSet,  Bestfeat, value, Sublabels)       return myTree                                       #决策树分类器, Inputtree is a decision tree, Fearlabel is the column type, Testvec is the vector to classify def classify (Inputtree,featlabels,testvec):       firststr = inputtree.keys () [0]      seconddict =  Inputtree[firststr]   &nbsP;  featindex = featlabels.index (FIRSTSTR)       key =  testVec[featIndex]      valueOfFeat = secondDict[key]       if isinstance (valueoffeat, dict):            classlabel = classify (Valueoffeat, featlabels, testvec)       else: classLabel = valueOfFeat       return classLabel     #序列化决策树def  storetree (inputtree,filename):       import pickle      fw = open ( FileName, ' W ')       pickle.dump (INPUTTREE,FW)        Fw.close ()        #提取决策树       def grabtree ( FileName):     &nbsP; import pickle      fr = open (filename)        return pickle.load (FR)

The input matrix is a table and the last column is the result.

ID3 algorithm is a rough algorithm, it can classify the symbolic variables, but it can't analyze the numerical data.

In addition, when selecting eigenvalues, it is preferred to choose more kinds of eigenvalues, so we need to improve the values.

Learning Log---Decision Tree algorithm ID3