简单数据挖掘ID3决策树分类预测python代码

以下是一个简单的基于ID3算法的决策树分类预测的Python代码：

import pandas as pd
import numpy as np
import math

# 定义计算熵的函数
def calc_entropy(data):
    label_col = data.iloc[:, -1]
    _, counts = np.unique(label_col, return_counts=True)
    probs = counts / len(label_col)
    entropy = sum(probs * -np.log2(probs))
    return entropy

# 定义计算信息增益的函数
def calc_info_gain(data, feature):
    entropy_before_split = calc_entropy(data)
    vals, counts = np.unique(data[feature], return_counts=True)
    probs = counts / sum(counts)
    entropy_after_split = 0
    for i in range(len(vals)):
        sub_data = data[data[feature] == vals[i]]
        entropy_after_split += probs[i] * calc_entropy(sub_data)
    info_gain = entropy_before_split - entropy_after_split
    return info_gain

# 定义获取最佳切分特征的函数
def get_best_split_feature(data):
    features = data.columns[:-1]
    best_feature = None
    best_info_gain = -1
    for feature in features:
        info_gain = calc_info_gain(data, feature)
        if info_gain > best_info_gain:
            best_info_gain = info_gain
            best_feature = feature
    return best_feature

# 定义决策树训练函数
def train_decision_tree(data):
    # 终止条件1：如果数据集中所有样本都属于同一类别，直接返回该类别
    if len(np.unique(data.iloc[:, -1])) == 1:
        return np.unique(data.iloc[:, -1])[0]
    # 终止条件2：如果数据集中没有特征可供切分，直接返回样本数最多的类别
    if len(data.columns) == 1:
        return np.bincount(data.iloc[:, -1]).argmax()
    # 选择最佳切分特征
    best_feature = get_best_split_feature(data)
    # 构建决策树
    decision_tree = {best_feature: {}}
    vals, counts = np.unique(data[best_feature], return_counts=True)
    for i in range(len(vals)):
        # 去掉最佳切分特征，递归构建子树
        sub_data = data[data[best_feature] == vals[i]].drop(best_feature, axis=1)
        decision_tree[best_feature][vals[i]] = train_decision_tree(sub_data)
    return decision_tree

# 定义决策树预测函数
def predict(data, tree):
    for feature, subtree in tree.items():
        val = data[feature]
        if val not in subtree:
            return np.bincount(data.iloc[:, -1]).argmax()
        subtree = subtree[val]
        if isinstance(subtree, dict):
            return predict(data.drop(feature, axis=1), subtree)
        else:
            return subtree

# 加载数据集
data = pd.read_csv('data.csv')

# 训练决策树
decision_tree = train_decision_tree(data)

# 预测新样本
new_data = pd.DataFrame({'feature1': [1, 1, 0, 0], 'feature2': [1, 0, 1, 0]})
for i in range(len(new_data)):
    prediction = predict(new_data.iloc[i], decision_tree)
    print('Sample', i+1, 'predicted class:', prediction)

注：此代码仅作为学习参考，实际应用中需要考虑更多的细节和优化措施。