decision tree

gistfile1.txt

'''
Created on Oct 12, 2010
Decision Tree Source Code for Machine Learning in Action Ch. 3
@author: Peter Harrington
'''
from math import log
import operator

# prepare data set
# create data set and corrsponding labels
def createDataSet():
    # yes: flippers
    # no: no surfacing
    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

# calculate shannon entropy
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:
        # calculate probability
        prob = float(labelCounts[key]) / numEntries
        # calculate shannon entropy
        shannonEnt -= prob * log(prob, 2)  # log base 2
    # the shannon entropy is a sum value
    return shannonEnt

# split data set by the feature[axis] and its feature value
# this method is just like a filter method and return the
def splitDataSet(dataSet, axis, value):
    retDataSet = []
    for featVec in dataSet:
        if featVec[axis] == value:
            # collect the left data set
            # the value is the selected feature and should be eliminate
            reducedFeatVec = featVec[:axis]  # chop out axis used for splitting
            # collect the right data set
            reducedFeatVec.extend(featVec[axis + 1:])
            retDataSet.append(reducedFeatVec)
    return retDataSet

# choose the best feature to split data set, return the best feature
def chooseBestFeatureToSplit(dataSet):
    numFeatures = len(dataSet[0]) - 1  # the last column is used for the labels
    # calculate the original shannon entropy
    baseEntropy = calcShannonEnt(dataSet)
    bestInfoGain = 0.0
    bestFeature = -1
    for i in range(numFeatures):  # iterate over all the features
        # collect the specific feature value
        # create a list of all the examples of this feature
        featList = [example[i] for example in dataSet]
        # get unique feature values
        uniqueVals = set(featList)  # get a set of unique values
        newEntropy = 0.0
        for value in uniqueVals:
            # split the data set using the specific feature and the
            # corresponding feature value
            subDataSet = splitDataSet(dataSet, i, value)
            prob = len(subDataSet) / float(len(dataSet))
            newEntropy += prob * calcShannonEnt(subDataSet)
        # calculate the changed value
        # calculate the info gain; ie reduction in entropy
        infoGain = baseEntropy - newEntropy
        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

# the algorithm of majority cnt
def majorityCnt(classList):
    # define a class counter collector
    classCount = {}
    for vote in classList:
        if vote not in classCount.keys():
            classCount[vote] = 0
        classCount[vote] += 1
    # iteritems is the iterator of the dict
    # operator is a another module support sorting function
    sortedClassCount = sorted(classCount.iteritems(),
                              key=operator.itemgetter(1), reverse=True)
    # return the majority classifier of the remain data set
    return sortedClassCount[0][0]

# create the decision tree
def createTree(dataSet, labels):
    # collect the labels as the classified classes
    classList = [example[-1] for example in dataSet]
    # case 1: all the data are belong to the the same class
    if classList.count(classList[0]) == len(classList):
        return classList[0]  # stop splitting when all of the classes are equal
    # case 2: there are no data to split any more. using the majority algorithm
    if len(dataSet[0]) == 1:  # stop splitting when there are no more features in dataSet
        return majorityCnt(classList)
    # case 3: running the algorithm
    # choose the best feature to split the data set
    bestFeat = chooseBestFeatureToSplit(dataSet)
    # get the best feature labels
    bestFeatLabel = labels[bestFeat]
    # build up the decision tree
    myTree = {bestFeatLabel: {}}
    # delete the best feature and its corresponding labels
    del(labels[bestFeat])
    # collect the feature values
    featValues = [example[bestFeat] for example in dataSet]
    # distinct the feature values
    uniqueVals = set(featValues)
    for value in uniqueVals:
        # simple deep copy
        # copy all of labels, so trees don't mess up existing labels
        subLabels = labels[:]
        # using the best feature and its value to split the data set
        # the split data set is a sub data set after split by the best feature and its values
        # with the sub labels correspondingly
        myTree[bestFeatLabel][value] = createTree(
            splitDataSet(dataSet, bestFeat, value), subLabels)
    return myTree

# using the tree classifier to classify the data set
def classify(inputTree, featLabels, testVec):
    firstStr = inputTree.keys()[0]
    secondDict = inputTree[firstStr]
    featIndex = featLabels.index(firstStr)
    key = testVec[featIndex]
    valueOfFeat = secondDict[key]
    # the decision node classify the testVec recursion
    if isinstance(valueOfFeat, dict):
        classLabel = classify(valueOfFeat, featLabels, testVec)
    # the leaf node classify the testVec as this label
    else:
        classLabel = valueOfFeat
    return classLabel

# store the classify tree persistence
def storeTree(inputTree, filename):
    import pickle
    fw = open(filename, 'w')
    pickle.dump(inputTree, fw)
    fw.close()

# load classify tree for the disk
def grabTree(filename):
    import pickle
    fr = open(filename)
    return pickle.load(fr)

decision tree code samples