python实现决策树ID3算法的示例代码

时间:2022-11-25 14:41:31

在周志华的西瓜书和李航的统计机器学习中对决策树ID3算法都有很详细的解释,如何实现呢?核心点有如下几个步骤

step1:计算香农熵

?
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
from math import log
import operator
 
 
# 计算香农熵
def calculate_entropy(data):
  label_counts = {}
  for feature_data in data:
    laber = feature_data[-1] # 最后一行是laber
    if laber not in label_counts.keys():
      label_counts[laber] = 0
    label_counts[laber] += 1
 
  count = len(data)
  entropy = 0.0
 
  for key in label_counts:
    prob = float(label_counts[key]) / count
    entropy -= prob * log(prob, 2)
  return entropy

step2.计算某个feature的信息增益的方法

?
1
2
3
4
5
6
7
8
9
10
11
12
13
# 计算某个feature的信息增益
# index:要计算信息增益的feature 对应的在data 的第几列
# data 的香农熵
def calculate_relative_entropy(data, index, entropy):
  feat_list = [number[index] for number in data] # 得到某个特征下所有值(某列)
  uniqual_vals = set(feat_list)
  new_entropy = 0
  for value in uniqual_vals:
    sub_data = split_data(data, index, value)
    prob = len(sub_data) / float(len(data))
    new_entropy += prob * calculate_entropy(sub_data) # 对各子集香农熵求和
  relative_entropy = entropy - new_entropy # 计算信息增益
  return relative_entropy

step3.选择最大信息增益的feature

?
1
2
3
4
5
6
7
8
9
10
11
12
13
14
# 选择最大信息增益的feature
def choose_max_relative_entropy(data):
  num_feature = len(data[0]) - 1
  base_entropy = calculate_entropy(data)#香农熵
  best_infor_gain = 0
  best_feature = -1
  for i in range(num_feature):
    info_gain=calculate_relative_entropy(data, i, base_entropy)
    #最大信息增益
    if (info_gain > best_infor_gain):
      best_infor_gain = info_gain
      best_feature = i
 
  return best_feature

step4.构建决策树

?
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
def create_decision_tree(data, labels):
  class_list=[example[-1] for example in data]
  # 类别相同,停止划分
  if class_list.count(class_list[-1]) == len(class_list):
    return class_list[-1]
  # 判断是否遍历完所有的特征时返回个数最多的类别
  if len(data[0]) == 1:
    return most_class(class_list)
  # 按照信息增益最高选取分类特征属性
  best_feat = choose_max_relative_entropy(data)
  best_feat_lable = labels[best_feat] # 该特征的label
  decision_tree = {best_feat_lable: {}} # 构建树的字典
  del(labels[best_feat]) # 从labels的list中删除该label
  feat_values = [example[best_feat] for example in data]
  unique_values = set(feat_values)
  for value in unique_values:
    sub_lables=labels[:]
    # 构建数据的子集合,并进行递归
    decision_tree[best_feat_lable][value] = create_decision_tree(split_data(data, best_feat, value), sub_lables)
  return decision_tree

在构建决策树的过程中会用到两个工具方法:

?
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
# 当遍历完所有的特征时返回个数最多的类别
def most_class(classList):
  class_count={}
  for vote in classList:
    if vote not in class_count.keys():class_count[vote]=0
    class_count[vote]+=1
  sorted_class_count=sorted(class_count.items,key=operator.itemgetter(1),reversed=True)
  return sorted_class_count[0][0]
  
# 工具函数输入三个变量(待划分的数据集,特征,分类值)返回不含划分特征的子集
def split_data(data, axis, value):
  ret_data=[]
  for feat_vec in data:
    if feat_vec[axis]==value :
      reduce_feat_vec=feat_vec[:axis]
      reduce_feat_vec.extend(feat_vec[axis+1:])
      ret_data.append(reduce_feat_vec)
  return ret_data

以上就是本文的全部内容,希望对大家的学习有所帮助,也希望大家多多支持服务器之家。

原文链接:https://segmentfault.com/a/1190000015083169