### 基于 ID3 算法的连续值决策树构建 为了实现基于 ID3 算法的连续值决策树，需要对连续特征进行离散化处理。以下是完整的实现方法，包括数据预处理、信息增益计算以及递归构建决策树的过程。 #### 数据预处理 对于连续特征，可以通过遍历所有可能的阈值来找到最佳划分点，从而将连续值转换为离散值[^4]。 ```python import numpy as np import pandas as pd from math import log2 # 定义西瓜数据集 data = { '色泽': ['青绿', '乌黑', '浅白', '青绿', '乌黑', '浅白', '青绿', '乌黑', '浅白'], '根蒂': ['蜷缩', '稍蜷', '硬挺', '蜷缩', '稍蜷', '硬挺', '蜷缩', '稍蜷', '硬挺'], '敲声': ['浊响', '沉闷', '清脆', '浊响', '沉闷', '清脆', '浊响', '沉闷', '清脆'], '纹理': ['清晰', '稍糊', '模糊', '清晰', '稍糊', '模糊', '清晰', '稍糊', '模糊'], '触感': ['软粘', '硬滑', '硬滑', '软粘', '硬滑', '硬滑', '软粘', '硬滑', '硬滑'], '含糖量': [0.3, 0.5, 0.7, 0.4, 0.6, 0.8, 0.35, 0.55, 0.75], '好瓜': ['是', '是', '否', '是', '否', '否', '是', '是', '否'] } df = pd.DataFrame(data) # 将连续特征离散化 def discretize_continuous_feature(X, y, feature): """ 找到最佳划分点并离散化连续特征 """ sorted_values = sorted(X[feature].unique()) thresholds = [(sorted_values[i] + sorted_values[i+1]) / 2 for i in range(len(sorted_values) - 1)] best_threshold = None best_info_gain = -float('inf') for threshold in thresholds: X_temp = X.copy() X_temp[feature] = X_temp[feature].apply(lambda x: '高' if x > threshold else '低') info_gain = information_gain(X_temp, y, feature) if info_gain > best_info_gain: best_info_gain = info_gain best_threshold = threshold X[feature] = X[feature].apply(lambda x: '高' if x > best_threshold else '低') return X # 对含糖量进行离散化 df = discretize_continuous_feature(df, df['好瓜'], '含糖量') ``` #### 信息增益计算 定义熵和信息增益的计算函数，用于评估每个特征的信息增益[^4]。 ```python def entropy(y): """ 计算熵 """ value_counts = y.value_counts(normalize=True) return -np.sum(value_counts * np.log2(value_counts)) def information_gain(X, y, feature): """ 计算信息增益 """ total_entropy = entropy(y) values = X[feature].unique() weighted_entropy = 0 for value in values: subset_y = y[X[feature] == value] weight = len(subset_y) / len(y) weighted_entropy += weight * entropy(subset_y) return total_entropy - weighted_entropy ``` #### 决策树构建 递归地选择信息增益最大的特征作为划分节点。 ```python def build_decision_tree(X, y, features): """ 构建决策树 """ if len(y.unique()) == 1: # 如果类别相同，则返回该类别 return y.iloc[0] if len(features) == 0: # 如果没有特征可选，则返回多数类别 return y.value_counts().idxmax() best_feature = max(features, key=lambda f: information_gain(X, y, f)) tree = {best_feature: {}} for value in X[best_feature].unique(): subset_X = X[X[best_feature] == value] subset_y = y[X[best_feature] == value] if len(subset_y) == 0: # 如果子集为空，则返回多数类别 tree[best_feature][value] = y.value_counts().idxmax() else: sub_features = [f for f in features if f != best_feature] tree[best_feature][value] = build_decision_tree(subset_X, subset_y, sub_features) return tree # 构建决策树 X = df.drop('好瓜', axis=1) y = df['好瓜'] features = X.columns.tolist() decision_tree = build_decision_tree(X, y, features) print(decision_tree) ``` #### 输出结果 最终生成的决策树将以嵌套字典的形式表示，便于后续可视化或预测使用。 ---