k近邻法matlab_【机器学习算法笔记】01-k近邻法

# 函数1 创建题设数据集
def create_movie_data_set():
    group_out = np.array([[1, 101], [5, 89], [108, 5], [115, 8]])
    labels_out = np.array(['爱情片', '爱情片', '动作片', '动作片'])
    return group_out, labels_out


# 函数2 读取鸾尾花数做实验（弃）
def create_iris_data_set(test_proportion=0.2):
    from sklearn import datasets
    from sklearn.model_selection import train_test_split
    iris_all = datasets.load_iris()
    iris_train, iris_test, iris_label_train, iris_label_test 
        = train_test_split(iris_all.data, iris_all.target,test_size=test_proportion)
    return iris_train, iris_test, iris_label_train, iris_label_test


# 函数3 读取书上约会数据实验
def create_dating_data_set(test_proportion=0.2):
    from sklearn.model_selection import train_test_split
    # 原始数据归一化处理
    dating_data = normalization1(np.loadtxt("datingTestSet.txt", delimiter='t', usecols=(0, 1, 2)))
    dating_target = np.loadtxt("datingTestSet.txt", delimiter='t', dtype = 'str',usecols=3)
    dating_train, dating_test, dating_label_train, dating_label_test 
        = train_test_split(dating_data, dating_target,test_size=test_proportion)
    return dating_train, dating_test, dating_label_train, dating_label_test

# 函数4 简单k邻近算法
def kNN1(x_in, data_train_in, label_train_in, k, n=2):
    """
    :param x_in: 待测试的数据点
    :param data_train_in: 训练集
    :param label_train_in: 训练集标签
    :param k: k近邻法参数
    :param n: 距离度量方法选择，默认n=2即计算欧氏距离
    :return label_result:测试点的标签
    """
    import numpy as np

    # 计算各点到目标点的距离，默认n=2即欧氏距离
    distance = (((data_train_in - x_in) ** n).sum(axis=1)) ** (1 / n)

    distance_sort = distance.argsort()  # 求取distance中从小到大的索引值

    label_dic = {}
    for i in range(k):
        # 读取距离最近的k个点的标签
        label_vote = label_train_in[distance_sort[i]]
        # 将该k个点的标签写入字典
        label_dic[label_vote] = label_dic.get(label_vote, 0) + 1
    # 读取最近k个点的标签值入字典后，直接求取字典中值最大的健即可
    label_result = max(label_dic.items(), key=lambda x: x[1])[0]  # 快速求取字典最值

    return label_result

# 函数5 简单直接归一化算法
def normalization1(array_in):
    array_in_max = array_in.max(0)
    array_in_min = array_in.min(0)
    array_out = (array_in-array_in_min)/(array_in_max - array_in_min)
    return array_out

if __name__ == '__main__':
    """直接验证尝试"""
       
    test_result = np.array([])
    for k_test in range(1,31):
        error_rate = np.array([])
        for n in range(200):

            data_train, data_test, label_train, label_test = create_iris_data_set(0.2)
            len_test = float(len(data_test))
            result_error = 0
            
            for i in range(len(data_test)):
                label_try = kNN1(data_test[i], data_train, label_train, k_test)
                # 检查统计每个输入实例分类结果是否与实际标签一致
                if label_try != label_test[i]:
                    result_error += 1
            # 统计每批的错误率
            error_rate = np.append(error_rate, float(result_error) / len_test)

        test_result = np.append(test_result,np.mean(error_rate))

    print(test_result)
    # 画图看一看
    plt.title("iris_data_kNN_test")
    plt.xlabel("k")
    plt.ylabel("error_rate")
    plt.plot(np.arange(1,31), test_result)
    plt.show()