简单决策树编程

设置好属性字典

D_keys = dict(色泽=['青绿', '乌黑', '浅白'], 根蒂=['蜷缩', '硬挺', '稍蜷'], 敲声=['清脆', '沉闷', '浊响'], 纹理=['稍糊', '模糊', '清晰'],
               脐部=['凹陷', '稍凹', '平坦'], 触感=['软粘', '硬滑'])

读取数据：

x = []
f = open(r'C:\Users\dell\Desktop\xiguaji3.0.txt', 'r')
lines = f.readlines()
for line in lines:
    line = line.replace('\n', '')
    x.append(line.split(","))
f.close()
x = np.array(x)
# print(x)

titles = list(x[0, 1:-3])  # 表头
y = x[:, -1]  # 标记
x = np.delete(x, 0, axis=0)     # 删除第一行，表头
x = np.delete(x, [0, 7, 8], axis=1)     # 删除第一

计算数据中的熵

# 重新定义log函数，针对底为0的情况
def log(x):
    if x != 0:
        return math.log(x, 2)
    else:
        return 0


# 计算titles中属性的熵
def culEnt(x,title):
    indexnumber = {}
    for m, n in enumerate(title):
        indexnumber[n] = {}
        for k in title[n]:
            indexnumber[n][k] = []
            for index, row in enumerate(x):
                if row[m] == k:
                    indexnumber[n][k].append(index)
    Ent = {}
    for key1 in indexnumber:
        a = 0.0
        for key2 in indexnumber[key1]:
            yes = 0
            no = 0
            if len(indexnumber[key1][key2]) != 0:
                for i in indexnumber[key1][key2]:
                    if x[i][-1] == '好瓜':
                        yes += 1
                    else:
                        no += 1
                p1 = float(yes / (yes + no))
                p2 = float((yes + no) / (x.shape[0]))
                a += p2 * (p1 * log(p1) + (1 - p1) * log((1 - p1)))  # 防止对0取对数，重新定义log函数
            else:
                a += 0
        Ent[key1] = -a

    return indexnumber, Ent

输出的indexnumber包含title中每个属性取值的样本编号；Ent包含每个属性的熵大小
输出结果如下：
{‘色泽’: {‘青绿’: [0, 3, 5, 9, 12, 16], ‘乌黑’: [1, 2, 6, 7, 8, 14], ‘浅白’: [4, 10, 11, 13, 15]},
‘根蒂’: {‘蜷缩’: [0, 1, 2, 3, 4, 11, 15, 16], ‘硬挺’: [9, 10], ‘稍蜷’: [5, 6, 7, 8, 12, 13, 14]},
‘敲声’: {‘清脆’: [9, 10], ‘沉闷’: [1, 3, 8, 13, 16], ‘浊响’: [0, 2, 4, 5, 6, 7, 11, 12, 14, 15]},
‘纹理’: {‘稍糊’: [6, 8, 12, 13, 16], ‘模糊’: [10, 11, 15], ‘清晰’: [0, 1, 2, 3, 4, 5, 7, 9, 14]},
#‘脐部’: {‘凹陷’: [0, 1, 2, 3, 4, 12, 13], ‘稍凹’: [5, 6, 7, 8, 14, 16], ‘平坦’: [9, 10, 11, 15]},
#‘触感’: {‘软粘’: [5, 6, 9, 11, 14], ‘硬滑’: [0, 1, 2, 3, 4, 7, 8, 10, 12, 13, 15, 16]}}

{‘色泽’: 0.88937738110375, ‘根蒂’: 0.8548275868023224, ‘敲声’: 0.8567211127541194,
‘纹理’: 0.6169106490008467, ‘脐部’: 0.7083437635274363, ‘触感’: 0.9914560571925497}

下面进行决策树递归编程，按照书上的算法：

def calculateShannonEnt(x, title, table):
    yes = 0                # 存储正例样本个数
    for row in x:
        if row[-1] == "好瓜":
            yes += 1
    p_yes = float(yes/x.shape[0])
    if p_yes == 1:
        print("leaf节点：好瓜")
        return                  # 结束递归
    if p_yes == 0:
        print("leaf节点：坏瓜")
        return
    judge = True
    for i in range(x.shape[1]):     # 列数，等于属性集
        if len(set(x[:, i])) != 1:  # 如果某一属性各样本取值存在不等的情况，则说明在属性集上不全相等
            judge = False
            break
    if len(title) == 0 | judge:  #len(title) == 0意味着属性集为空
    	print("leaf节点：好瓜") if p_yes >= 0.5 else print("leaf节点：坏瓜")
        return

    indexnumber, Ent = culEnt(x, title)  # indexnumber：每个属性包含的样本情况，Ent：每个属性的信息增益
    # print(indexnumber)

    key_name = min(Ent, key=Ent.get)   # 最优划分属性

    print("节点:{0}：分支{1}".format(key_name, indexnumber[key_name]))
    
    for key in indexnumber[key_name]:       
        if len(indexnumber[key_name][key]) == 0:
            print("leaf节点：好瓜") if p_yes >= 0.5 else print("leaf节点：坏瓜")
       		return
        else:
            print("分支:", key)
            m = []              #存储去除最优属性后的新的属性
            for i in table:
                if i != key_name:
                    m.append(i)
            n = {}          #存储去除最优属性后的新的属性字典
            for i in title:
                if i != key_name:
                    n[i] = title[i]

            a = np.array(x[indexnumber[key_name][key]])
            a=np.delete(a, table.index(key_name), axis=1) #划分后的样本
        
            calculateShannonEnt(a, n, m)

最后输出结果与书上一致
节点:纹理：分支{‘稍糊’: [6, 8, 12, 13, 16], ‘模糊’: [10, 11, 15], ‘清晰’: [0, 1, 2, 3, 4, 5, 7, 9, 14]}
分支: 稍糊
节点:触感：分支{‘软粘’: [0], ‘硬滑’: [1, 2, 3, 4]}
分支: 软粘
leaf节点：好瓜
分支: 硬滑
leaf节点：坏瓜
分支: 模糊
leaf节点：坏瓜
分支: 清晰
节点:根蒂：分支{‘蜷缩’: [0, 1, 2, 3, 4], ‘硬挺’: [7], ‘稍蜷’: [5, 6, 8]}
分支: 蜷缩
leaf节点：好瓜
分支: 硬挺
leaf节点：坏瓜
分支: 稍蜷
节点:色泽：分支{‘青绿’: [0], ‘乌黑’: [1, 2], ‘浅白’: []}
分支: 青绿
leaf节点：好瓜
分支: 乌黑
节点:触感：分支{‘软粘’: [1], ‘硬滑’: [0]}
分支: 软粘
leaf节点：坏瓜
分支: 硬滑
leaf节点：好瓜
leaf节点:好瓜 浅白