更新 function.py

function.py

@@ -1,6 +1,7 @@
 import numpy as np
 from sklearn.impute import SimpleImputer
-from sklearn.feature_selection import SelectKBest, f_classif
+from sklearn.feature_selection import SelectKBest, mutual_info_classif
+from sklearn.preprocessing import RobustScaler
 
 # 填写学号和姓名
 id = "202210423213"
@@ -39,36 +40,26 @@ def load_data(filepath, k_features=20):
     labels = np.array(labels)
 
     # 使用SimpleImputer填充NaN
-    imputer = SimpleImputer(strategy='mean')
+    imputer = SimpleImputer(strategy='median')
     data = imputer.fit_transform(data)
 
-    # 特征标准化，避免除零错误
-    mean = data.mean(axis=0)
-    std_dev = data.std(axis=0)
-    std_dev[std_dev == 0] = 1  # 如果标准差为0，将其设为1，避免除以零
-    data = (data - mean) / std_dev
+    # 使用RobustScaler进行数据标准化，减少异常值的影响
+    scaler = RobustScaler()
+    data = scaler.fit_transform(data)
 
-    # 特征选择
-    selector = SelectKBest(f_classif, k=k_features)
+    # 特征选择，使用互信息
+    selector = SelectKBest(mutual_info_classif, k=k_features)
     data = selector.fit_transform(data, labels)
 
     return data, labels
 
 def sigmoid(z):
-    """
-    Sigmoid激活函数，用于对率回归模型。
-
-    Args:
-    - z (numpy array): 输入数据。
-
-    Returns:
-    - numpy array: Sigmoid函数输出值。
-    """
+    """Sigmoid激活函数"""
     return 1 / (1 + np.exp(-z))
 
-def train_logreg(X, y, initial_lr=0.005, iterations=5000, regularization=0.01, decay=0.0001):
+def train_logreg(X, y, initial_lr=0.01, iterations=5000, regularization=0.02, decay=0.00005, patience=200):
     """
-    对率回归训练函数，包含学习率衰减策略。
+    对率回归训练函数，包含学习率衰减和早停机制。
 
     Args:
     - X (numpy array): 特征数据。
@@ -77,23 +68,36 @@ def train_logreg(X, y, initial_lr=0.005, iterations=5000, regularization=0.01, d
     - iterations (int): 迭代次数。
     - regularization (float): 正则化强度。
     - decay (float): 学习率衰减系数。
+    - patience (int): 早停的等待次数。
 
     Returns:
     - numpy array: 训练得到的权重。
     """
     m, n = X.shape
     weights = np.zeros(n)
-    learning_rate = initial_lr
+    best_loss = float('inf')
+    stop_counter = 0
 
     for i in range(iterations):
+        # 前向传播和损失计算
         z = np.dot(X, weights)
         predictions = sigmoid(z)
+        loss = -np.mean(y * np.log(predictions + 1e-9) + (1 - y) * np.log(1 - predictions + 1e-9)) + regularization * np.sum(weights ** 2) / (2 * m)
+        
+        # 早停机制
+        if loss < best_loss:
+            best_loss = loss
+            stop_counter = 0
+        else:
+            stop_counter += 1
+            if stop_counter >= patience:
+                print("早停：在第 {} 轮迭代停止".format(i))
+                break
+        
+        # 反向传播更新权重
         errors = y - predictions
         gradient = np.dot(X.T, errors) / m - regularization * weights / m
-        weights += learning_rate * gradient
-
-        # 学习率衰减
-        learning_rate = initial_lr / (1 + decay * i)
+        weights += initial_lr * gradient / (1 + decay * i)  # 学习率衰减
 
     return weights
 
@@ -119,7 +123,7 @@ if __name__ == "__main__":
     X_test, y_test = load_data("./data/horse-colic.test", k_features=20)
 
     # 训练模型
-    weights = train_logreg(X_train, y_train, initial_lr=0.005, iterations=5000, regularization=0.01, decay=0.0001)
+    weights = train_logreg(X_train, y_train, initial_lr=0.01, iterations=5000, regularization=0.02, decay=0.00005, patience=200)
 
     # 模型评估
     correct_predictions = 0
@@ -130,4 +134,4 @@ if __name__ == "__main__":
         print(f"预测结果: {y_pred}, 实际结果: {y_test[i]}")
 
     accuracy = correct_predictions / len(X_test)
-    print(f"测试集准确率: {accuracy:.2f}")
+    print(f"测试集准确率: {accuracy:.2f}")