更新 function.py

function.py

@@ -28,140 +28,106 @@ name = "万子玉"
 
 # 这里写自己的代码实现
 def sigmoid(z):
-    return 1 / (1 + np.exp(-z))
+    return 1.0 / (1 + np.exp(-z))
 
-def train_logreg(X, y, learning_rate=0.1, iterations=2000, lambda_reg=0.01, batch_size=32):
+def compute_gradient(X_batch, y_batch, weights, lambda_reg):
+    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  # L2正则化
+    return gradient
+
+def train_logreg(X, y, learning_rate=0.001, iterations=10000, batch_size=64, lambda_reg=0.1):
     m, n = X.shape
-    weights = np.random.randn(n) * 0.01  # 随机初始化权重，取较小值
+    weights = np.random.randn(n) * np.sqrt(2. / n)  # He初始化
+    best_weights = weights.copy()
+    best_accuracy = 0
 
-    for i in range(iterations):
-        # 每次迭代打乱数据
+    for _ in range(iterations):
         indices = np.random.permutation(m)
-        X = X[indices]
-        y = y[indices]
+        X_shuffled = X[indices]
+        y_shuffled = y[indices]
 
-        for j in range(0, m, batch_size):  # 按照 batch size 迭代
-            X_batch = X[j:j + batch_size]
-            y_batch = y[j:j + batch_size]
-
-            z = np.dot(X_batch, weights)
-            predictions = sigmoid(z)
-            gradient = np.dot(X_batch.T, (predictions - y_batch)) / X_batch.shape[0]
-
-            # L2正则化
-            gradient += (lambda_reg / m) * weights
-
-            # 更新权重
+        for j in range(0, m, batch_size):
+            X_batch = X_shuffled[j:j + batch_size]
+            y_batch = y_shuffled[j:j + batch_size]
+            gradient = compute_gradient(X_batch, y_batch, weights, lambda_reg)
             weights -= learning_rate * gradient
 
-        # 动态调整学习率，逐渐减小
-        if i % 100 == 0:  # 每100次迭代输出一次学习率
-            print(f"Iteration {i}, Learning Rate: {learning_rate}")
+        # 计算训练集准确率
+        train_accuracy = accuracy(X, y, weights)
+        if train_accuracy > best_accuracy:
+            best_accuracy = train_accuracy
+            best_weights = weights.copy()
 
-    return weights
+        # 动态调整学习率
+        if _ % 100 == 0 and train_accuracy > 0.95:  # 达到高准确率时减小学习率
+            learning_rate *= 0.9
 
-    """
-    对率回归训练函数。
+    return best_weights
 
-    Args:
-    - X (numpy array): 特征数据。
-    - y (numpy array): 标签数据。
-    - other_params (str): 其它参数，可选设置。若需设置其它参数，必须设置其默认值，否则将**无法通过**自动评测，**成绩为零**。
+def predict_logreg(X, weights):
+    z = np.dot(X, weights)
+    h = sigmoid(z)
+    return np.where(h >= 0.5, 1, 0)
 
-    Returns:
-    - numpy array: 训练得到的权重。
-    """
-
-
-def preprocess_data(X):
-    """特征标准化。"""
-    scaler = StandardScaler()
-    return scaler.fit_transform(X)
-
-    """
-    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。
-    """
-
-# 这里也写自己的代码实现
 def load_data(filepath):
-    """加载数据并处理缺失值。"""
-    data = []
-    labels = []
-
+    data, labels = [], []
     with open(filepath, 'r') as f:
         for line in f:
-            values = line.strip().replace('?', '0').split()
-            try:
-                values = [float(v) for v in values]
-                
-                # 合并特征23的取值
-	                if values[22] == 2 or values[22] == 3:  # 特征23
-                    labels.append(0)  # 死亡
-                elif values[22] == 1:
-                    labels.append(1)  # 生存
-                else:
-                    continue  # 无效标签，丢弃
-                
-                # 用0填充缺失值
-                values = [0 if v == 0 else v for v in values]  # 直接用0填充
-                data.append(values[:2] + values[3:23])  # 保留特征1, 2 和特征4-23
-            except ValueError:
-                continue
+            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]))
 
     data = np.array(data)
     labels = np.array(labels)
 
-    return data, labels
-    """
-    从文件中加载数据，若不自行正确实现，则无法通过评测，**不记录成绩**。
+    # 检查特征数量
+    num_features = data.shape[1]
 
-    Args:
-    - directory (str): 数据文件路径。
+	    # 特征选择：去掉特征3（索引2）和特征24~28（索引23~26），确保索引有效
+    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（索引23到26）
 
-    Returns:
-    - np.array: 特征数据。
-    - np.array: 标签数据。
-    """
+    # 删除特征
+    data = np.delete(data, indices_to_delete, axis=1)
 
+	    # 将标签二分类化，特征23（索引22）取值2和3合并为0，其余为1
+    target_variable = np.where((labels == 2) | (labels == 3), 0, 1)
+
+    return data, target_variable
+
+def accuracy(X, y, weights):
+    predictions = predict_logreg(X, weights)
+    return np.mean(predictions == y)
+
+# 其余代码保持不变
 if __name__ == "__main__":
-    # 加载数据集
+    # 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)
 
-    # 归一化数据
-    X = preprocess_data(X)
+    # 数据归一化
+    scaler = MinMaxScaler()
+    X_train = scaler.fit_transform(X_train)
+    X_val = scaler.transform(X_val)
 
-    # 使用 train_test_split 打乱并随机抽取数据
-    X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42)
+    # 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)
 
-    # K折交叉验证
-    k = 5
-    kf = KFold(n_splits=k, shuffle=True, random_state=42)
-    accuracies = []
+    # 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}")
 
-    for train_index, val_index in kf.split(X_train):
-        X_train_kf, X_val_kf = X_train[train_index], X_train[val_index]
-        y_train_kf, y_val_kf = y_train[train_index], y_train[val_index]
+    # 加载测试集
+    X_test, y_test = load_data("./data/horse-colic.test")
+    X_test = scaler.transform(X_test)
 
-        # 训练逻辑回归模型，设置 batch size
-        batch_size = 32
-        weights = train_logreg(X_train_kf, y_train_kf, learning_rate=0.1, iterations=2000, lambda_reg=0.01, batch_size=batch_size)
-
-        # 验证模型
-        val_predictions = [predict_logreg(X_val_kf[i], weights) for i in range(len(X_val_kf))]
-        accuracy = np.mean(val_predictions == y_val_kf)
-        accuracies.append(accuracy)
-
-    # 输出交叉验证的准确率和平均准确率
-    print(f"K折交叉验证得分: {accuracies}")
-    print(f"平均准确率: {np.mean(accuracies)}")
-
-    # 测试单个数据
-    y_pred = predict_logreg(X_test[11], weights)
-    print(f"预测结果: {y_pred}, 实际结果: {y_test[11]}")
+    # 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}")