基于_统计学习方法_的机器学习代码复现.zip

""" 数据集：play_tennis 五折五次交叉验证 每折正确率最高为100% 平均正确率为76.66% 由于数据量过小，划分出的决策树会出现未包含所有特征的情况，故对未知特征采取多数表决法确定标签 """ import pandas as pd import numpy as np from sklearn.model_selection import KFold class DecisionTree: def __init__(self): self.feature = "" # 表示待划分特征 self.children = {} # 表示子树，key为具体特征值，value为对应子树 self.amount = 0 # 表示当前样本数量，用于多数投票表决确定未知特征的标签 def Entropy(data): # 对列向量计算经验熵 items = list(set(data)) p = np.zeros(len(items)) size = data.size for i in range(len(items)): p[i] = np.sum(data == items[i])/size return np.sum(-1 * np.log2(p) * p) def Conditional_E(data): H_Y_X = np.zeros(data.shape[1] - 1) i = 0 for col in data.columns[:-1]: # 最后一列不取 column = data.loc[:, col] # 遍历每一列，当前列存为column items = list(set(column)) e = np.zeros(len(items)) p = np.zeros_like(e) for it in range(len(items)): e[it] = Entropy(data[column == items[it]].iloc[:, -1]) # 根据当前列是否等于items特征来选取data，再取最后一列计算经验熵 p[it] = data[column == items[it]].shape[0]/data.shape[0] H_Y_X[i] = e @ p i += 1 return H_Y_X def CreatTree(data, epsilon, tree): if len(data.columns) == 1: # 如果数据只有一列（标签列），则说明所有特征均被划分 target = data.iloc[:, -1].value_counts() # 统计结果，取最多数为结果 tree.feature = target.index[0] return H = Entropy(data.iloc[:, -1]) H_Y_X = Conditional_E(data) Gain = H - H_Y_X index = np.argmax(Gain) if Gain[index] < epsilon: target = data.iloc[:, -1].value_counts() # 统计结果，取最多数为结果 tree.feature = target.index[0] tree.amount = data.shape[0] return feature = data.columns[index] # 确定划分特征 tree.feature = feature tree.amount = data.shape[0] tree.children = {item: DecisionTree() for item in list(set(data.iloc[:, index]))} for key, value in tree.children.items(): CreatTree(data[data.loc[:, feature] == key].drop(feature, axis=1), epsilon, value) def Predict(x, tree): try: next_tree = tree.children[x[tree.feature]] except KeyError: # 此时x的待划分特征不在决策树中 vote = {'Yes': 0, 'No': 0} # 统计投票结果 Vote(tree, vote) result = 'Yes' if vote['Yes'] >= vote['No'] else 'No' return result if not next_tree.children: return next_tree.feature result = Predict(x, next_tree) return result def Vote(tree, vote_result): # 当此时决策树中划分特征中出现未知值，则遍历该所有子树，统计样本数量，投票表决含未知值的样本标签 for node in tree.children.values(): if not node.children: vote_result[node.feature] += node.amount continue Vote(node, vote_result) df = pd.read_csv("play_tennis.csv", index_col="day") kf = KFold(n_splits=5, shuffle=True, random_state=1) accuracy = np.zeros(5) it = 0 for train_index, test_index in kf.split(df): train = df.iloc[train_index, :] root = DecisionTree() CreatTree(train, 0.1, root) for test in test_index: if Predict(df.iloc[test, :], root) == df.iloc[test, -1]: accuracy[it] += 1 accuracy[it] /= len(test_index) print('正确率为', accuracy[it] * 100, '%') it += 1 print('平均正确率为', np.sum(accuracy) / 5 * 100, '%')