# 导入相关模块import numpy as npfrom collections import Counter
import matplotlib.pyplot as pltfrom sklearn import datasets
from sklearn.utils import shuffledef load_data():iris = datasets.load_iris()# 打乱数据后的数据和标签X, y = shuffle(iris.data, iris.target, random_state=13)# 数据转换为 flout32 格式X = X.astype(np.float32)# 简单划分训练集和测试集, 训练样本 - 测试样本比例为 7:3offset = int(X.shape[0] * 0.7)X_train, y_train = X[:offset], y[:offset]X_test, y_test = X[offset:], y[offset:]# 将标签转换为竖向量y_train = y_train.reshape((-1, 1))y_test = y_test.reshape((-1, 1))return X_train, X_test, y_train, y_testdef compute_distances(X, X_train):"""定义欧氏距离函数X: 测试样本实例矩阵X_train: 训练样本实例矩阵"""# 测试实例样本num_test = X.shape[0]# 训练实例样本量num_train = X_train.shape[0]# 基于训练和测试维度的欧氏距离初始化dists = np.zeros((num_test, num_train))# 测试样本鱼训练样本的矩阵点乘M = np.dot(X, X_train.T)# 测试样本矩阵平方te = np.square(X).sum(axis=1)# 训练样本矩阵平方tr = np.square(X_train).sum(axis=1)# 计算欧式距离dists = np.sqrt(-2 * M + tr + np.matrix(te).T)return distsdef predict_labels(y_train, dists, k=1):"""定义预测函数:param y_train: 训练集标签:param dists: 测试集与训练集之间的欧氏距离矩阵:param k: k值:return: 测试集预测结果"""# 测试样本量num_test = dists.shape[0]# 初始化测试集预测结果y_pred = np.zeros(num_test)# 遍历for i in range(num_test):# 初始化最近邻列表closest_y = []# 按 欧式距离矩阵排序后取索引, 并用训练集标签按排序后的索引取值# 最后展平列表# 注意 np.argsort 函数的用法labels = y_train[np.argsort(dists[i, :])].flatten()# 取最近的k个值closest_y = labels[0:k]# 对最近的k个值进行计数统计# 这里注意 collections 模块中的计数器 Counter 的用法c = Counter(closest_y)# 取计数最多的那个类别y_pred[i] = c.most_common(1)[0][0]return y_predif __name__ == '__main__':# 导入 sklearn iris 数据集X_train, X_test, y_train, y_test = load_data()dists = compute_distances(X=X_test, X_train=X_train)y_test_pred = predict_labels(y_train=y_train, dists=dists, k=1)y_test_pred = y_test_pred.reshape((-1, 1))# 找出预测正确的实例num_correct = np.sum(y_test_pred == y_test)# 计算分类准确率accuracy = float(num_correct) / X_test.shape[0]print('KNN Accuracy based on NumPy: ' + str(accuracy))# 用五折交叉验证寻找最优 k值# 五折num_folds = 5# 候选 k 值k_choices = [1, 3, 5, 8, 10, 12, 15, 20, 50, 100]X_train_folds = []y_train_folds = []# 训练数据划分X_train_folds = np.array_split(X_train, num_folds)# 训练标签划分y_train_folds = np.array_split(y_train, num_folds)k_to_accuracies = {}# 表里所有候选k值for k in k_choices:# 五折遍历for fold in range(num_folds):# 为 传入的训练集单独划分出一个验证集作为测试集validation_X_test = X_train_folds[fold]validation_y_test = y_train_folds[fold]temp_X_train = np.concatenate(X_train_folds[:fold] + X_train_folds[fold+1:])temp_y_train = np.concatenate(y_train_folds[:fold] + y_train_folds[fold+1:])# 计算距离temp_dists = compute_distances(X=validation_X_test, X_train=temp_X_train)temp_y_test_pred = predict_labels(temp_y_train, temp_dists, k=k)temp_y_test_pred = temp_y_test_pred.reshape((-1, 1))# 查看分类准确率num_correct = np.sum(temp_y_test_pred == validation_y_test)num_test = validation_X_test.shape[0]accuracy = float(num_correct) / num_testk_to_accuracies[k] = k_to_accuracies.get(k, []) + [accuracy]for k in sorted(k_to_accuracies):for accuracy in k_to_accuracies[k]:print(f'k = {k}, accuracy = {accuracy}')# 打印不同k值, 不同折数下的分类准确率for k in k_choices:# 取出第k个k值的分类准确率accuracies = k_to_accuracies[k]# 绘制不同k值下分类准确率的散点图plt.scatter([k] * len(accuracies), accuracies)# 计算分类准确率均值并排序accuracies_mean = np.array([np.mean(v) for k, v in sorted(k_to_accuracies.items())])# 计算分类准确率标准差并排序accuracies_std = np.array([np.std(v) for k, v in sorted(k_to_accuracies.items())])# 绘制有质询区间的误差棒图plt.errorbar(k_choices, accuracies_mean, yerr=accuracies_std)# 绘图标题plt.title('Cross-validation on k')# x轴标签plt.xlabel('k')# y轴标签plt.ylabel('Cross-validation accuracy')plt.show()

if __name__ == '__main__':# 导入 KNeighborsClassifier 模块from sklearn.neighbors import KNeighborsClassifier# 创建 k近邻实例neigh = KNeighborsClassifier(n_neighbors=10)# k 近邻模型拟合neigh.fit(X_train, y_train)# k 近邻模型预测y_pred = neigh.predict(X_test)# 预测结果数组重塑y_pred = y_pred.reshape((-1, 1))# 统计预测正确的个数num_correct = np.sum(y_pred == y_test)# 计算分类准确率accuracy = float(num_correct) / X_test.shape[0]print(f'KNN Accuracy based on sklearn: {accuracy}.')