更新 function.py

function.py

@@ -31,111 +31,88 @@ name = "严锡好"
 
 
 # 这里写自己的代码实现
-# Sigmoid函数，用于逻辑回归模型
 def sigmoid(z):
     return 1.0 / (1 + np.exp(-z))
 
-# 计算梯度，用于逻辑回归的梯度下降算法
 def compute_gradient(X_batch, y_batch, weights, lambda_reg):
-    z = np.dot(X_batch, weights)  # 计算z值
-    h = sigmoid(z)  # 计算sigmoid函数值
-    gradient = np.dot(X_batch.T, (h - y_batch)) / len(y_batch)  # 计算梯度
-    gradient += (lambda_reg / len(y_batch)) * weights  # 添加L2正则化项
+    z = np.dot(X_batch, weights)
+    h = sigmoid(z)
+    gradient = np.dot(X_batch.T, (h - y_batch)) / len(y_batch)
+    gradient += (lambda_reg / len(y_batch)) * weights
     return gradient
 
-# 训练逻辑回归模型
+# Adam优化器更新函数
+def adam_update(grad, m, v, t, lr=0.001, beta1=0.9, beta2=0.999, epsilon=1e-8):
+    m = beta1 * m + (1 - beta1) * grad
+    v = beta2 * v + (1 - beta2) * (grad ** 2)
+    m_hat = m / (1 - beta1 ** t)
+    v_hat = v / (1 - beta2 ** t)
+    return lr * m_hat / (np.sqrt(v_hat) + epsilon), m, v
+
 def train_logreg(X, y, learning_rate=0.001, iterations=10000, batch_size=64, lambda_reg=0.1):
-    m, n = X.shape  # 获取样本数量和特征数量
-    weights = np.random.randn(n) * np.sqrt(2. / n)  # 使用He初始化权重
-    best_weights = weights.copy()  # 存储最佳权重
-    best_accuracy = 0  # 存储最佳准确率
+    m, n = X.shape
+    weights = np.random.randn(n) * np.sqrt(2. / n)
+    m_t, v_t = np.zeros(n), np.zeros(n)
+    best_weights = weights.copy()
+    best_accuracy = 0
 
-    for _ in range(iterations):  # 迭代指定次数
-        indices = np.random.permutation(m)  # 随机打乱数据
-        X_shuffled = X[indices]  # 打乱特征
-        y_shuffled = y[indices]  # 打乱标签
+    for epoch in range(iterations):
+        indices = np.random.permutation(m)
+        X_shuffled, y_shuffled = X[indices], y[indices]
 
-        for j in range(0, m, batch_size):  # 按批次处理数据
-            X_batch = X_shuffled[j:j + batch_size]  # 获取当前批次的特征
-            y_batch = y_shuffled[j:j + batch_size]  # 获取当前批次的标签
-            gradient = compute_gradient(X_batch, y_batch, weights, lambda_reg)  # 计算梯度
-            weights -= learning_rate * gradient  # 更新权重
+        for j in range(0, m, batch_size):
+            X_batch, y_batch = X_shuffled[j:j + batch_size], y_shuffled[j:j + batch_size]
+            grad = compute_gradient(X_batch, y_batch, weights, lambda_reg)
+            update, m_t, v_t = adam_update(grad, m_t, v_t, epoch + 1, learning_rate)
+            weights -= update
 
-        # 计算训练集准确率
         train_accuracy = accuracy(X, y, weights)
-        if train_accuracy > best_accuracy:  # 如果当前准确率更高，则更新最佳权重和准确率
+        if train_accuracy > best_accuracy:
             best_accuracy = train_accuracy
             best_weights = weights.copy()
 
-        # 动态调整学习率
-        if _ % 100 == 0 and train_accuracy > 0.95:  # 每100次迭代检查一次，如果准确率超过0.95，则减小学习率
+        if epoch % 100 == 0 and train_accuracy > 0.95:
             learning_rate *= 0.9
 
     return best_weights
 
-# 预测函数，用于逻辑回归模型
 def predict_logreg(X, weights):
-    z = np.dot(X, weights)  # 计算z值
-    h = sigmoid(z)  # 计算sigmoid函数值
-    return np.where(h >= 0.5, 1, 0)  # 返回预测结果
+    z = np.dot(X, weights)
+    h = sigmoid(z)
+    return np.where(h >= 0.5, 1, 0)
 
-# 加载数据函数
 def load_data(filepath):
     data, labels = [], []
-    with open(filepath, 'r') as f:  # 打开文件
-        for line in f:  # 逐行读取
-            line = line.strip().split()  # 去除空格并分割
-            if len(line) >= 1 and line[-1].isdigit():  # 确保行数据有效
-                line_data = [0 if x == '?' else float(x) for x in line[:-1]]  # 将非数字替换为0并转换为浮点数
+    with open(filepath, 'r') as f:
+        for line in f:
+            line = line.strip().split()
+            if len(line) >= 1 and line[-1].isdigit():
+                line_data = [0 if x == '?' else float(x) for x in line[:-1]]
                 data.append(line_data)
-                labels.append(int(line[-1]))  # 将标签转换为整数
+                labels.append(int(line[-1]))
 
-    data = np.array(data)  # 转换为numpy数组
-    labels = np.array(labels)
-
-    # 检查特征数量并进行特征选择
-    num_features = data.shape[1]
-    indices_to_delete = [2]  # 特征3
-    if num_features > 24:  # 确保特征数量足够
-        indices_to_delete += list(range(24, 27))  # 特征24到27
-
-    # 删除特征
+    data, labels = np.array(data), np.array(labels)
+    indices_to_delete = [2] + (list(range(24, 27)) if data.shape[1] > 24 else [])
     data = np.delete(data, indices_to_delete, axis=1)
-
-    # 将标签二分类化
     target_variable = np.where((labels == 2) | (labels == 3), 0, 1)
-
     return data, target_variable
 
-# 计算准确率函数
 def accuracy(X, y, weights):
-    predictions = predict_logreg(X, weights)  # 预测结果
-    return np.mean(predictions == y)  # 计算准确率
+    predictions = predict_logreg(X, weights)
+    return np.mean(predictions == y)
 
-# 主函数
 if __name__ == "__main__":
-    # 加载数据
     X, y = load_data("./data/horse-colic.data")
-
-    # 拆分训练集和验证集
     X_train, X_val, y_train, y_val = train_test_split(X, y, test_size=0.2, random_state=42)
 
-    # 数据归一化
     scaler = MinMaxScaler()
-    X_train = scaler.fit_transform(X_train)
-    X_val = scaler.transform(X_val)
+    X_train, X_val = scaler.fit_transform(X_train), scaler.transform(X_val)
 
-    # 训练模型
     final_weights = train_logreg(X_train, y_train, learning_rate=0.01, iterations=5000, batch_size=64, lambda_reg=0.1)
-
-    # 验证模型
     val_accuracy = accuracy(X_val, y_val, final_weights)
     print(f"验证集的准确率: {val_accuracy:.4f}")
 
-    # 加载测试集
     X_test, y_test = load_data("./data/horse-colic.test")
     X_test = scaler.transform(X_test)
-
-    # 测试模型
     final_test_accuracy = accuracy(X_test, y_test, final_weights)
     print(f"最终测试集的准确率: {final_test_accuracy:.4f}")