更新 function.py

function.py

@@ -29,33 +29,21 @@ def load_data(filepath):
     - X (numpy array): 包含非线性特征的特征数据
     - y (numpy array): 标签数据
     """
-    # 读取数据，指定分隔符为空格，缺失值符号为 '?'
     data = np.genfromtxt(filepath, delimiter=" ", dtype=str)
-    
-    # 将 '?' 替换为 np.nan 并转换为浮点数
     data[data == '?'] = np.nan
     data = data.astype(float)
-    
-    # 对每一列的缺失值使用列的均值填充
     col_means = np.nanmean(data, axis=0)
     inds = np.where(np.isnan(data))
     data[inds] = np.take(col_means, inds[1])
-    
-    # 分离特征和标签
     X = data[:, :-1]
     y = data[:, -1]
-    
-    # 生成非线性特征 (平方和对数)
     square_features = X ** 2
-    log_features = np.log1p(X)  # 使用 log1p 避免 log(0)
-    
-    # 将非线性特征添加到原特征集
+    log_features = np.log1p(X)
     X = np.hstack([X, square_features, log_features])
-    
     return X, y
 
 # 对率回归训练函数
-def train_logreg(X, y, learning_rate=0.01, num_iterations=1000, regularization_strength=0):
+def train_logreg(X, y, initial_learning_rate=0.1, num_iterations=2000, regularization_strength=0.1):
     """
     对率回归训练函数。
 
@@ -69,79 +57,50 @@ def train_logreg(X, y, learning_rate=0.01, num_iterations=1000, regularization_s
     Returns:
     - weights (numpy array): 训练得到的权重。
     """
+   
     m, n = X.shape
-    weights = np.zeros(n)
-    for _ in range(num_iterations):
+    weights = np.random.randn(n) * 0.01  # 随机初始化权重
+    learning_rate = initial_learning_rate
+    for i in range(num_iterations):
         predictions = 1 / (1 + np.exp(-np.dot(X, weights)))
         gradient = (1 / m) * np.dot(X.T, (predictions - y)) + regularization_strength * weights
         weights -= learning_rate * gradient
+        if i % 100 == 0:  # 每100次迭代降低学习率
+            learning_rate *= 0.9
     return weights
 
 # 决策树训练函数
 def train_decision_tree(X, y):
-    """
-    训练简单的决策树模型。
-
-    Args:
-    - X (numpy array): 特征数据
-    - y (numpy array): 标签数据
-
-    Returns:
-    - clf (DecisionTreeClassifier): 训练好的决策树模型
-    """
     clf = DecisionTreeClassifier(max_depth=5, random_state=42)
     clf.fit(X, y)
     return clf
 
 # 对率回归预测函数
 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。
-    """
     probability = 1 / (1 + np.exp(-np.dot(inX, weights)))
-    return 1 if probability >= 0.4 else 0
+    return 1 if probability >= 0.5 else 0
+
 
 # 融合预测函数
-def predict_fusion(inX, weights, clf):
-    """
-    融合对率回归和决策树的预测结果。
-
-    Args:
-    - inX (numpy array): 输入待分类的样本。
-    - weights (numpy array): 对率回归的权重。
-    - clf (DecisionTreeClassifier): 训练好的决策树模型。
-
-    Returns:
-    - int: 最终预测的分类标签。
-    """
-    # 对率回归预测
-    prediction_logreg = predict_logreg(inX, weights)
-
-    # 决策树预测
-    prediction_tree = clf.predict(inX.reshape(1, -1))[0]
-    
-    # 融合投票
-    final_prediction = 1 if (prediction_logreg + prediction_tree) >= 1 else 0
-    return final_prediction
+def predict_fusion(inX, weights, clf, alpha=0.6):
+    prob_logreg = 1 / (1 + np.exp(-np.dot(inX, weights)))
+    prob_tree = clf.predict_proba(inX.reshape(1, -1))[0, 1]
+    final_prob = alpha * prob_logreg + (1 - alpha) * prob_tree
+    return 1 if final_prob >= 0.5 else 0
 
 if __name__ == "__main__":
-    # 以下代码可用于测试，可自行随意修改
-    # 加载训练集
     X_train, y_train = load_data("./data/horse-colic.data")
-
-    # 加载测试集
     X_test, y_test = load_data("./data/horse-colic.test")
 
     # 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, initial_learning_rate=0.1, num_iterations=2000)
+
+    # 训练决策树模型
+    clf = train_decision_tree(X_train, y_train)
 
     # 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]}")
+    y_pred_logreg = predict_logreg(X_test[11], weights)
+    y_pred_fusion = predict_fusion(X_test[11], weights, clf)
+
+    print(f"对率回归预测结果: {y_pred_logreg}, 实际结果: {y_test[11]}")
+    print(f"融合预测结果: {y_pred_fusion}, 实际结果: {y_test[11]}")