CART决策树可视化实现（纯代码）

CART决策树可视化实现

目录

1、CART算法实现代码

# ssy_1_CART决策树
import operator
import matplotlib.pylab as plt
import matplotlib


# 能够显示中文
matplotlib.rcParams['font.sans-serif'] = ['SimHei']
matplotlib.rcParams['font.serif'] = ['SimHei']

def createDataset():
    # 数据集D
    dataSet = [['青绿', '蜷缩', '浊响', '清晰', '凹陷', '硬滑', '好瓜'],
               ['乌黑', '蜷缩', '沉闷', '清晰', '凹陷', '硬滑', '好瓜'],
               ['乌黑', '蜷缩', '浊响', '清晰', '凹陷', '硬滑', '好瓜'],
               ['青绿', '蜷缩', '沉闷', '清晰', '凹陷', '硬滑', '好瓜'],
               ['浅白', '蜷缩', '浊响', '清晰', '凹陷', '硬滑', '好瓜'],
               ['青绿', '稍蜷', '浊响', '清晰', '稍凹', '软粘', '好瓜'],
               ['乌黑', '稍蜷', '浊响', '稍糊', '稍凹', '软粘', '好瓜'],
               ['乌黑', '稍蜷', '浊响', '清晰', '稍凹', '硬滑', '好瓜'],
               ['乌黑', '稍蜷', '沉闷', '稍糊', '稍凹', '硬滑', '坏瓜'],
               ['青绿', '硬挺', '清脆', '清晰', '平坦', '软粘', '坏瓜'],
               ['浅白', '硬挺', '清脆', '模糊', '平坦', '硬滑', '坏瓜'],
               ['浅白', '蜷缩', '浊响', '模糊', '平坦', '软粘', '坏瓜'],
               ['青绿', '稍蜷', '浊响', '稍糊', '凹陷', '硬滑', '坏瓜'],
               ['浅白', '稍蜷', '沉闷', '稍糊', '凹陷', '硬滑', '坏瓜'],
               ['乌黑', '稍蜷', '浊响', '清晰', '稍凹', '软粘', '坏瓜'],
               ['浅白', '蜷缩', '浊响', '模糊', '平坦', '硬滑', '坏瓜'],
               ['青绿', '蜷缩', '沉闷', '稍糊', '稍凹', '硬滑', '坏瓜']]

    # 属性集A
    labels = ['色泽', '根蒂', '敲声', '纹理', '脐部', '触感']
    return dataSet, labels

def majorityCnt(classList):
    classCount = {}

    for vote in classList:
        if vote not in classCount.keys():
            classCount[vote] = 0
        classCount[vote] += 1

    # 降序
    sortedClassCount = sorted(classCount.items(), key=operator.itemgetter(1), reverse=True)
    print(type(sortedClassCount))
    print(sortedClassCount)
    return sortedClassCount[0][0]

def calcGini(dataSet):
    numEntries = len(dataSet)
    labelCounts = {}

    for featVec in dataSet:
        currentLabel = featVec[-1]
        if currentLabel not in labelCounts.keys():
            labelCounts[currentLabel] = 0
        labelCounts[currentLabel] += 1

    for key in labelCounts:
        labelCounts[key] /= numEntries
        labelCounts[key] = labelCounts[key] * labelCounts[key]

    Gini = 1 - sum(labelCounts.values())
    return Gini

def splitDataSet(dataSet, axis, value):
    retDataSet1 = []
    retDataSet2 = []
    for featVec in dataSet:
        if featVec[axis] == value:
            reducedFeatVec = featVec[:axis]
            reducedFeatVec.extend(featVec[axis + 1:])
            retDataSet1.append(reducedFeatVec)
        else:
            reducedFeatVec = featVec[:axis]
            reducedFeatVec.extend(featVec[axis + 1:])
            retDataSet2.append(reducedFeatVec)

    return retDataSet1, retDataSet2

def chooseBestFeatureToSplit(dataSet):
    numFeatures = len(dataSet[0]) - 1
    if numFeatures == 0:
        return 0
    bestGini = 1
    bestFeature = -1

    for i in range(numFeatures):
        featList = [example[i] for example in dataSet]
        uniqueVals = set(featList)
        Gini = {}

        for value in uniqueVals:
            subDataSet1, subDataSet2 = splitDataSet(dataSet, i, value)
            prob1 = len(subDataSet1) / float(len(dataSet))
            prob2 = len(subDataSet2) / float(len(dataSet))
            subDataSet1Gini = calcGini(subDataSet1)
            subDataSet2Gini = calcGini(subDataSet2)
            Gini[value] = prob1 * subDataSet1Gini + prob2 * subDataSet2Gini

            if Gini[value] < bestGini:
                bestGini = Gini[value]
                bestFeature = i
                bestSplit = value

    return bestFeature, bestSplit

def createTree(dataSet, labels):
    # 获得每一个标签
    classList = [example[-1] for example in dataSet]

    # 标签全相同即全属于同一类别，返回该标签
    if classList.count(classList[0]) == len(dataSet):
        return classList[0]
    # 所有样本在所有属性上取值相同，类别标记为样本数最多的类
    if len(dataSet[0]) == 1:
        return majorityCnt(classList)

    # 获取最优索引
    bestFeat, bestSplit = chooseBestFeatureToSplit(dataSet)
    # 获取最优索引的名称
    bestFeatLabel = labels[bestFeat]

    # 创建根节点
    myTree = {bestFeatLabel: {}}
    # 删除用过的结点
    del (labels[bestFeat])
    subLabels = labels[:]
    subDataSet1, subDataSet2 = splitDataSet(dataSet, bestFeat, bestSplit)
    myTree[bestFeatLabel][bestSplit] = createTree(subDataSet1, subLabels)
    myTree[bestFeatLabel]['others'] = createTree(subDataSet2, subLabels)

    return myTree

2、可视化代码

# 代码可视化部分

# 分叉节点，也就是决策节点
decisionNode = dict(boxstyle="sawtooth", fc="0.8")

# 叶子节点
leafNode = dict(boxstyle="round4", fc="0.8")

# 箭头样式
arrow_args = dict(arrowstyle="<-")

def plotNode(nodeTxt, centerPt, parentPt, nodeType):
    """
    绘制一个节点
    :param nodeTxt: 描述该节点的文本信息
    :param centerPt: 文本的坐标
    :param parentPt: 点的坐标，这里也是指父节点的坐标
    :param nodeType: 节点类型,分为叶子节点和决策节点
    :return:
    """
    createPlot.ax1.annotate(nodeTxt, xy=parentPt, xycoords='axes fraction',
                            xytext=centerPt, textcoords='axes fraction',
                            va="center", ha="center", bbox=nodeType, arrowprops=arrow_args)

def getNumLeafs(myTree):
    """
    获取叶节点的数目
    :param myTree:
    :return:
    """
    # 统计叶子节点的总数
    numLeafs = 0

    # 得到当前第一个key，也就是根节点
    firstStr = list(myTree.keys())[0]

    # 得到第一个key对应的内容
    secondDict = myTree[firstStr]

    # 递归遍历叶子节点
    for key in secondDict.keys():
        # 如果key对应的是一个字典，就递归调用
        if type(secondDict[key]).__name__ == 'dict':
            numLeafs += getNumLeafs(secondDict[key])
        # 不是的话，说明此时是一个叶子节点
        else:
            numLeafs += 1
    return numLeafs

def getTreeDepth(myTree):
    """
    得到数的深度层数
    :param myTree:
    :return:
    """
    # 用来保存最大层数
    maxDepth = 0

    # 得到根节点
    firstStr = list(myTree.keys())[0]

    # 得到key对应的内容
    secondDic = myTree[firstStr]

    # 遍历所有子节点
    for key in secondDic.keys():
        # 如果该节点是字典，就递归调用
        if type(secondDic[key]).__name__ == 'dict':
            # 子节点的深度加1
            thisDepth = 1 + getTreeDepth(secondDic[key])

        # 说明此时是叶子节点
        else:
            thisDepth = 1

        # 替换最大层数
        if thisDepth > maxDepth:
            maxDepth = thisDepth

    return maxDepth

def plotMidText(cntrPt, parentPt, txtString):
    """
    计算出父节点和子节点的中间位置，填充信息
    :param cntrPt: 子节点坐标
    :param parentPt: 父节点坐标
    :param txtString: 填充的文本信息
    :return:
    """
    # 计算x轴的中间位置
    xMid = (parentPt[0] - cntrPt[0]) / 2.0 + cntrPt[0]
    # 计算y轴的中间位置
    yMid = (parentPt[1] - cntrPt[1]) / 2.0 + cntrPt[1]
    # 进行绘制
    createPlot.ax1.text(xMid, yMid, txtString)

def plotTree(myTree, parentPt, nodeTxt):
    """
    绘制出树的所有节点，递归绘制
    :param myTree: 树
    :param parentPt: 父节点的坐标
    :param nodeTxt: 节点的文本信息
    :return:
    """
    # 计算叶子节点数
    numLeafs = getNumLeafs(myTree=myTree)

    # 计算树的深度
    depth = getTreeDepth(myTree=myTree)

    # 得到根节点的信息内容
    firstStr = list(myTree.keys())[0]

    # 计算出当前根节点在所有子节点的中间坐标,也就是当前x轴的偏移量加上计算出来的根节点的中心位置作为x轴（比如说第一次：初始的x偏移量为：-1/2W,计算出来的根节点中心位置为：(1+W)/2W，相加得到：1/2），当前y轴偏移量作为y轴
    cntrPt = (plotTree.xOff + (1.0 + float(numLeafs)) / 2.0 / plotTree.totalW, plotTree.yOff)

    # 绘制该节点与父节点的联系
    plotMidText(cntrPt, parentPt, nodeTxt)

    # 绘制该节点
    plotNode(firstStr, cntrPt, parentPt, decisionNode)

    # 得到当前根节点对应的子树
    secondDict = myTree[firstStr]

    # 计算出新的y轴偏移量，向下移动1/D，也就是下一层的绘制y轴
    plotTree.yOff = plotTree.yOff - 1.0 / plotTree.totalD

    # 循环遍历所有的key
    for key in secondDict.keys():
        # 如果当前的key是字典的话，代表还有子树，则递归遍历
        if isinstance(secondDict[key], dict):
            plotTree(secondDict[key], cntrPt, str(key))
        else:
            # 计算新的x轴偏移量，也就是下个叶子绘制的x轴坐标向右移动了1/W
            plotTree.xOff = plotTree.xOff + 1.0 / plotTree.totalW
            # 打开注释可以观察叶子节点的坐标变化
            # print((plotTree.xOff, plotTree.yOff), secondDict[key])
            # 绘制叶子节点
            plotNode(secondDict[key], (plotTree.xOff, plotTree.yOff), cntrPt, leafNode)
            # 绘制叶子节点和父节点的中间连线内容
            plotMidText((plotTree.xOff, plotTree.yOff), cntrPt, str(key))

    # 返回递归之前，需要将y轴的偏移量增加，向上移动1/D，也就是返回去绘制上一层的y轴
    plotTree.yOff = plotTree.yOff + 1.0 / plotTree.totalD

def createPlot(inTree):
    """
    需要绘制的决策树
    :param inTree: 决策树字典
    :return:
    """
    # 创建一个图像
    fig = plt.figure(1, facecolor='white')
    fig.clf()
    axprops = dict(xticks=[], yticks=[])
    createPlot.ax1 = plt.subplot(111, frameon=False, **axprops)
    # 计算出决策树的总宽度
    plotTree.totalW = float(getNumLeafs(inTree))
    # 计算出决策树的总深度
    plotTree.totalD = float(getTreeDepth(inTree))
    # 初始的x轴偏移量，也就是-1/2W，每次向右移动1/W，也就是第一个叶子节点绘制的x坐标为：1/2W，第二个：3/2W，第三个：5/2W，最后一个：(W-1)/2W
    plotTree.xOff = -0.5 / plotTree.totalW
    # 初始的y轴偏移量，每次向下或者向上移动1/D
    plotTree.yOff = 1.0
    # 调用函数进行绘制节点图像
    plotTree(inTree, (0.5, 1.0), '')
    # 绘制
    plt.show()

3、实现

dataSet, labels = createDataset()
myTree = createTree(dataSet, labels)
print(myTree)
createPlot(myTree)

4、结果截图