From f3636cd5ae05e0fdd605aec7b7d5d30ad9459b79 Mon Sep 17 00:00:00 2001 From: cy <939601402@qq.com> Date: Sun, 10 Nov 2024 23:46:59 +0800 Subject: [PATCH] =?UTF-8?q?=E6=9B=B4=E6=96=B0=20function.py?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit --- function.py | 161 ++++++++++++++++++++++++---------------------------- 1 file changed, 74 insertions(+), 87 deletions(-) diff --git a/function.py b/function.py index e3d6fa2..0412122 100644 --- a/function.py +++ b/function.py @@ -1,6 +1,7 @@ +# function.py import numpy as np from sklearn.preprocessing import StandardScaler -from sklearn.model_selection import KFold +from itertools import product """ 实验说明: @@ -24,78 +25,86 @@ from sklearn.model_selection import KFold id = "202210423117" name = "程悦" + def sigmoid(z): - """计算Sigmoid激活函数值""" return 1.0 / (1 + np.exp(-z)) -def train_logreg(X, y, learning_rate=0.05, iterations=1500, batch_size=64, - regularization='L2', lambda_reg=0.05, cv=5): - """训练逻辑回归模型并使用交叉验证计算权重""" +def train_logreg(X, y, learning_rate=0.05, iterations=1500, batch_size=64, regularization='L2', lambda_reg=0.05): + """ + 对率回归训练函数。 + + Args: + - X (numpy array): 特征数据。 + - y (numpy array): 标签数据。 + + Returns: + - numpy array: 训练得到的权重。 + """ m, n = X.shape - kf = KFold(n_splits=cv, shuffle=True) - weights_list = [] - fold_scores = [] + weights = np.random.randn(n) * 0.01 # 随机初始化权重 + for i in range(iterations): + indices = np.random.permutation(m) # 随机打乱数据 + X_shuffled = X[indices] + y_shuffled = y[indices] - for train_index, val_index in kf.split(X): - X_train, X_val = X[train_index], X[val_index] - y_train, y_val = y[train_index], y[val_index] - weights = np.random.randn(n) * 0.01 + for j in range(0, m, batch_size): + X_batch = X_shuffled[j:j + batch_size] + y_batch = y_shuffled[j:j + batch_size] + z = np.dot(X_batch, weights) + h = sigmoid(z) + gradient = np.dot(X_batch.T, (h - y_batch)) / len(y_batch) - for _ in range(iterations): - indices = np.random.permutation(len(X_train)) - X_shuffled, y_shuffled = X_train[indices], y_train[indices] + # 添加正则化项 + if regularization == 'L1': + gradient += lambda_reg * np.sign(weights) / len(y_batch) + elif regularization == 'L2': + gradient += lambda_reg * weights / len(y_batch) - for j in range(0, len(X_train), batch_size): - X_batch = X_shuffled[j:j + batch_size] - y_batch = y_shuffled[j:j + batch_size] - z = np.dot(X_batch, weights) - h = sigmoid(z) - gradient = np.dot(X_batch.T, (h - y_batch)) / len(y_batch) + weights -= learning_rate * gradient - # 正则化处理 - if regularization == 'L1': - gradient += lambda_reg * np.sign(weights) / len(y_batch) - elif regularization == 'L2': - gradient += (lambda_reg / len(y_train)) * weights + # 可选：学习率衰减 + if i % 100 == 0: + learning_rate *= 0.95 # 每100次迭代衰减5% - weights -= learning_rate * gradient - - if _ % 100 == 0: - learning_rate *= 0.95 # 动态调整学习率 - - # 验证集准确率 - predictions = predict_logreg(X_val, weights) - score = np.mean(predictions == y_val) - fold_scores.append(score) - weights_list.append(weights) - - # 权重加权平均 - total_score = sum(fold_scores) - if total_score == 0: - return weights_list[-1] - - averaged_weights = np.zeros(n) - for weight, score in zip(weights_list, fold_scores): - averaged_weights += weight * score - return averaged_weights / total_score + return weights -def predict_logreg(X, weights): - """对输入数据进行逻辑回归预测""" - z = np.dot(X, weights) +def predict_logreg(inX, weights): + """ + 对率回归预测函数。 + + Args: + - inX (numpy array): 输入待分类的样本。 + - weights (numpy array): 训练得到的权重。 + + Returns: + - int: 预测的分类标签。 + """ + z = np.dot(inX, weights) h = sigmoid(z) - return (h > 0.5).astype(int) + return 1 if h > 0.5 else 0 def load_data(filepath): - """加载数据并处理缺失值""" - data, labels = [], [] + """ + 从文件中加载数据，填充缺失值，并进行特征处理。 + Args: + - filepath (str): 数据文件路径。 + + Returns: + - np.array: 特征数据。 + - np.array: 标签数据。 + """ + data = [] + labels = [] with open(filepath, 'r') as f: - for line in f: + lines = f.readlines() + for line in lines: line = line.strip().split() - if line and line[-1].isdigit(): + if len(line) >= 1 and line[-1].isdigit(): + # 将缺失值 '?' 填充为 0，并转换为浮点数 line_data = [1] + [0 if x == '?' else float(x) for x in line[:-1]] data.append(line_data) labels.append(int(line[-1])) @@ -103,47 +112,25 @@ def load_data(filepath): data = np.array(data) labels = np.array(labels) - # 处理目标变量 - labels[(labels == 2) | (labels == 3)] = 0 - data = np.delete(data, [2] + list(range(23, 28)), axis=1) + # 选择特征，合并目标变量 + target_variable = labels + target_variable[(target_variable == 2) | (target_variable == 3)] = 0 # 合并死亡和安乐死为0 + + # 去掉特征3和特征24~28 + data = np.delete(data, [2] + list(range(23, 28)), axis=1) # 特征3的索引是2，特征24-28的索引是23-27 - # 数据标准化 scaler = StandardScaler() data = scaler.fit_transform(data) - return data, labels - - -def accuracy(X, y, weights): - """计算模型的准确率""" - predictions = predict_logreg(X, weights) - return np.mean(predictions == y) - - -def ensemble_predict(X, models, weights): - """集成模型预测""" - predictions = np.array([predict_logreg(X, model) for model in models]) # (num_models, num_samples) - weighted_predictions = np.tensordot(predictions, weights, axes=(0, 0)) # 权重加权 - return (weighted_predictions > 0.5).astype(int) # 返回最终预测 - + return data, target_variable if __name__ == "__main__": X_train, y_train = load_data("./data/horse-colic.data") X_test, y_test = load_data("./data/horse-colic.test") - # 训练多个模型 - num_models = 5 - models = [train_logreg(X_train, y_train, learning_rate=0.02, iterations=6000, - batch_size=48, regularization='None', lambda_reg=0.02) - for _ in range(num_models)] + final_weights = train_logreg(X_train, y_train, learning_rate=0.05, iterations=1500, batch_size=64, regularization='L2', lambda_reg=0.05) - # 模型准确率作为权重 - model_accuracies = [accuracy(X_train, y_train, model) for model in models] - weights = np.array(model_accuracies) / sum(model_accuracies) - - # 集成预测 - ensemble_predictions = ensemble_predict(X_test, models, weights) - - final_accuracy = np.mean(ensemble_predictions == y_test) - print(f"最终测试集的准确率: {final_accuracy:.4f}") + # 计算并输出最终测试集的准确率 + final_accuracy = accuracy(X_test, y_test, final_weights) + print(f"最终测试集的准确率: {final_accuracy:.4f}") \ No newline at end of file