更新 function.py

function.py

@@ -3,6 +3,7 @@ import numpy as np
 from collections import Counter
 from scipy.spatial import distance
 from sklearn.preprocessing import StandardScaler
+from sklearn.model_selection import KFold
 
 """
 挑战说明:
@@ -41,7 +42,7 @@ name = "程悦"
 
 
 # 这里写自己的代码实现
-def classify0(inX, dataSet, labels, k=9):
+def classify0(inX, dataSet, labels, inv_cov_matrix, k=9):
     """
     KNN 分类算法。
 
@@ -54,20 +55,15 @@ def classify0(inX, dataSet, labels, k=9):
     Returns:
     - int: 预测的分类标签。
     """
-    # 计算协方差矩阵并求逆矩阵
-    cov_matrix = np.cov(dataSet, rowvar=False)
-    inv_cov_matrix = np.linalg.inv(cov_matrix)
-
     # 计算输入样本与训练集样本的马氏距离
-    distances = [distance.mahalanobis(inX, data_point, inv_cov_matrix) for data_point in dataSet]
+    distances = distance.cdist(dataSet, np.array([inX]), 'mahalanobis', VI=inv_cov_matrix).flatten()
 
     # 按距离排序并选择前 k 个最近的邻居
-    k_indices = np.argsort(distances)[:k]
-    k_nearest_labels = [labels[i] for i in k_indices]
+    k_indices = np.argpartition(distances, k)[:k]
+    k_nearest_labels = labels[k_indices]
 
     # 统计最近邻居中最多的标签
-    label_counts = Counter(k_nearest_labels)
-    most_common_label = label_counts.most_common(1)[0][0]
+    most_common_label = Counter(k_nearest_labels).most_common(1)[0][0]
 
     return most_common_label
 
@@ -80,7 +76,34 @@ def load_data(file_path):
     features = data[:, :-1]
     labels = data[:, -1].astype(int)
     return features, labels
+# 交叉验证
+def cross_validate(train_data, train_labels, k_values, n_splits=8):
+    kf = KFold(n_splits=n_splits)
+    accuracies = {k: [] for k in k_values}
 
+    # 计算协方差矩阵并求逆矩阵
+    cov_matrix = np.cov(train_data, rowvar=False)
+    inv_cov_matrix = np.linalg.inv(cov_matrix)
+
+    for k in k_values:
+        for train_index, val_index in kf.split(train_data):
+            X_train, X_val = train_data[train_index], train_data[val_index]
+            y_train, y_val = train_labels[train_index], train_labels[val_index]
+
+            # 预测验证集
+            correct_count = sum(
+                classify0(X_val[i], X_train, y_train, inv_cov_matrix, k) == y_val[i]
+                for i in range(len(X_val))
+            )
+
+            accuracy = correct_count / len(X_val)
+            accuracies[k].append(accuracy)
+
+    # 计算每个 k 值的平均准确率
+    avg_accuracies = {k: np.mean(v) for k, v in accuracies.items()}
+    best_k = max(avg_accuracies, key=avg_accuracies.get)
+
+    return best_k, avg_accuracies
 
 if __name__ == "__main__":
     # 以下代码可用于测试，可自行随意修改
@@ -90,21 +113,24 @@ if __name__ == "__main__":
     train_data, train_labels = load_data(train_file)
     test_data, test_labels = load_data(test_file)
 
-    # # 特征缩放
-    # scaler = StandardScaler()
-    # train_data = scaler.fit_transform(train_data)
-    # test_data = scaler.transform(test_data)
+    # 特征缩放
+    scaler = StandardScaler()
+    train_data = scaler.fit_transform(train_data)
+    test_data = scaler.transform(test_data)
 
-    # 测试 classify0 函数
-    test_sample = test_data[0]  # 使用测试集中的第一个样本进行测试
-    predicted_label = classify0(test_sample, train_data, train_labels)
-    print(f"Predicted label: {predicted_label}, Actual label: {test_labels[0]}")
+    # 使用交叉验证选择最佳的 k 值
+    k_values = range(1, 21)  # 尝试的 k 值范围
+    best_k, avg_accuracies = cross_validate(train_data, train_labels, k_values)
 
-    # 遍历整个测试集来评估分类器的性能
-    correct_count = 0
-    for i in range(len(test_data)):
-        predicted_label = classify0(test_data[i], train_data, train_labels)
-        if predicted_label == test_labels[i]:
-            correct_count += 1
+    print(f"Best k: {best_k}")
+    print("Average accuracies for each k:", avg_accuracies)
+
+    # 使用最佳 k 值测试分类器性能
+    correct_count = sum(
+        classify0(test_data[i], train_data, train_labels, np.linalg.inv(np.cov(train_data, rowvar=False)), best_k) ==
+        test_labels[i]
+        for i in range(len(test_data))
+    )
     accuracy = correct_count / len(test_data)
-    print(f"Accuracy: {accuracy * 100:.2f}%")
+    print(f"Accuracy with best k ({best_k}): {accuracy * 100:.2f}%")
+