更新 function.py

function.py

@@ -17,12 +17,14 @@ def compute_gradient(X_batch, y_batch, weights, lambda_reg):
     gradient += (lambda_reg / len(y_batch)) * weights  # L2正则化
     return gradient
 
-def train_logreg(X, y, learning_rate=0.001, iterations=10000, batch_size=64, lambda_reg=0.1):
+def train_logreg(X, y, learning_rate=0.01, iterations=20000, batch_size=64, lambda_reg=0.01, patience=5):
     """对率回归训练函数。"""
     m, n = X.shape
     weights = np.random.randn(n) * np.sqrt(2. / n)  # He初始化
-    print("开始训练对率回归模型")
-    
+    best_weights = weights.copy()
+    best_accuracy = 0
+    wait = 0
+
     for i in range(iterations):
         indices = np.random.permutation(m)
         X_shuffled = X[indices]
@@ -33,11 +35,20 @@ def train_logreg(X, y, learning_rate=0.001, iterations=10000, batch_size=64, lam
             y_batch = y_shuffled[j:j + batch_size]
             gradient = compute_gradient(X_batch, y_batch, weights, lambda_reg)
             weights -= learning_rate * gradient
-            
-        if i % 1000 == 0:  # 每1000次迭代打印一次信息
-            print(f"迭代 {i}: 当前权重形状: {weights.shape}")
 
-    return weights
+        # 早停法：在验证集上评估模型性能
+        accuracy = np.mean([1 if sigmoid(np.dot(X[i], weights)) >= 0.5 and y[i] == 1 or sigmoid(np.dot(X[i], weights)) < 0.5 and y[i] == 0 else 0 for i in range(m)])
+        if accuracy > best_accuracy:
+            best_accuracy = accuracy
+            best_weights = weights.copy()
+            wait = 0
+        else:
+            wait += 1
+            if wait >= patience:
+                print(f"Early stopping at iteration {i+1}")
+                break
+
+    return best_weights
 
 def predict_logreg(inX, weights):
     """对率回归预测函数。"""
@@ -81,7 +92,7 @@ def load_data(filepath):
     print(f"Data loaded successfully, data shape: {data.shape}, label shape: {target_variable.shape}")
     return data, target_variable
 
-def k_fold_cross_validation(X, y, k=5, learning_rate=0.001, iterations=10000, batch_size=64, lambda_reg=0.1):
+def k_fold_cross_validation(X, y, k=5, learning_rate=0.01, iterations=20000, batch_size=64, lambda_reg=0.01):
     """K折交叉验证。"""
     m = X.shape[0]
     fold_size = m // k
@@ -117,7 +128,7 @@ if __name__ == "__main__":
     X_train, mean, std = standardize_data(X_train)
 
     # K折交叉验证
-    avg_accuracy = k_fold_cross_validation(X_train, y_train, k=5, learning_rate=0.01, iterations=1000)
+    avg_accuracy = k_fold_cross_validation(X_train, y_train, k=5, learning_rate=0.01, iterations=20000)
     print(f"Average accuracy of K-fold cross-validation: {avg_accuracy:.4f}")
 
     # 加载测试集
@@ -128,10 +139,10 @@ if __name__ == "__main__":
         exit()
 
     # 测试集也需要进行相同的标准化
-    X_test = (X_test - mean) / std  # 使用训练集的均值和标准差进行标准化
+    X_test = (X_test - mean) / std
 
     # 训练对率回归模型
-    weights = train_logreg(X_train, y_train, learning_rate=0.01, iterations=1000)
+    weights = train_logreg(X_train, y_train, learning_rate=0.01, iterations=20000)
 
     # 测试单个数据
     y_pred = predict_logreg(X_test[11], weights)