用python实现id3算法

好的，下面是一个简单的Python实现ID3算法的例子：

import pandas as pd
import numpy as np
import math

class Node:
    def __init__(self, feature=None, label=None, child=None):
        self.feature = feature
        self.label = label
        self.child = child

def entropy(data):
    target = data[:, -1]
    _, count = np.unique(target, return_counts=True)
    p = count / count.sum()
    return -sum(p * np.log2(p))

def split_data(data, feature_index):
    feature = data[:, feature_index]
    values, count = np.unique(feature, return_counts=True)
    res = []
    for value in values:
        sub_data = data[feature == value]
        res.append(sub_data)
    return res

def info_gain(data, feature_index):
    feature = data[:, feature_index]
    h_d = entropy(data)
    h_d_a = 0
    for sub_data in split_data(data, feature_index):
        h_d_a += (sub_data.shape[0]/data.shape[0]) * entropy(sub_data)
    return h_d - h_d_a

def majority_label(data):
    target = data[:, -1]
    values, count = np.unique(target, return_counts=True)
    max_index = count.argmax()
    return values[max_index]

def build_tree(data, features):
    if len(np.unique(data[:, -1])) == 1:
        label = data[0, -1]
        return Node(label=label)

    if len(features) == 0:
        label = majority_label(data)
        return Node(label=label)

    info_gain_list = [info_gain(data, feature_index) for feature_index in features]
    max_index = np.array(info_gain_list).argmax()
    max_feature = features[max_index]

    root = Node(feature=max_feature)

    for sub_data in split_data(data, max_feature):
        child = build_tree(sub_data, np.delete(features, max_index))
        root.child.append(child)

    return root

def predict(root, x):
    if root.label is not None:
        return root.label

    for child in root.child:
        if child.feature == x[root.feature]:
            return predict(child, x)

if __name__ == '__main__':
    data = pd.read_csv('data.csv').values
    features = [i for i in range(data.shape[1]-1)]
    root = build_tree(data, features)
    x = np.array([1, 'S'])
    print(predict(root, x))

这里使用了pandas读取数据，需要将数据存为csv文件。其中Node类表示决策树的节点，包含特征、标签和子节点。entropy函数计算数据集的熵，split_data函数根据特征划分数据集，info_gain函数计算信息增益，majority_label函数返回数据集中出现最多的标签，build_tree函数递归构建决策树，predict函数预测新的数据的标签。