python实现KNN分类算法

K近邻算法，简称KNN算法，作用就是分类
三大要素

• 距离度量（常用利用欧式距离和曼哈顿距离比较远近）
• K值选择（最近的K个邻居）
• 决策规则（少数服从多数）
  
  已知训练集
  

import numpy
import matplotlib.pyplot as plt

'''
已知训练集和训练集类别、测试集
1.测试集广播和训练集一样的规格
2.计算距离
3.对距离列表排序返回最近的K个点的下标
4.有几个类别就设几个标签用来统计，循环排序列表，对类别判断，少数服从多数
5.数据可视化
'''

if __name__ == '__main__':
    pass
    k = 5 # 已知  奇数
    trainLabel = ["B", "B", "A", "A", "B", "B", "A", "B", "A", "A", "B", "A"]   # 训练集标签类别
    traindata = numpy.loadtxt("knndata1.txt", delimiter=",")    # 训练集中的坐标数据
    testdata = numpy.array([0.7, 0.7])  # 测试点

    # print(traindata)
    # print(testdata)

    # 测试点广播成和训练集相同的规格
    testdata = numpy.tile(testdata, (traindata.shape[0], 1))    # 不广播也行，为了画图方便还是广播
    # print(testdata)

    # 计算距离
    manDist = numpy.sum(abs(testdata - traindata), axis=1)  # 曼哈顿
    # print(manDist)
    eucDist = numpy.sum((testdata - traindata)**2, axis=1) ** 0.5  # 欧式距离  一般用欧式距离
    # print(eucDist)

    # 排序
    sortIndex = numpy.argsort(eucDist)
    # sortIndex = numpy.argsort(manDist)
    # print(sortIndex)

    a = b = 0
    for i in sortIndex[0:k]:
        if trainLabel[i] == "A":
            a += 1
        else:
            b += 1

    print(a, b)

    print("I am A") if a>b else print("I am B")


    # 数据可视化
    plt.figure()
    plt.title("WGS")
    for i in range(traindata.shape[0]): # 有几个点循环几次
        if trainLabel[i] == "A":
            plt.scatter(traindata[i, 0], traindata[i, 1], c="r")
        else:
            plt.scatter(traindata[i, 0], traindata[i, 1], c="g")
    if a>b:
        plt.scatter(testdata[0, 0], testdata[0, 1], c="r", marker="*", label="Test point")
    else:
        plt.scatter(testdata[0, 0], testdata[0, 1], c="g", marker="*", label="Test point")
    plt.grid(True)
    plt.legend(bbox_to_anchor=(0, 1.1), loc=2, borderaxespad=0)     # https://blog.csdn.net/Poul_henry/article/details/82533569
    # plt.show()

小案例

import numpy
import matplotlib.pyplot as plt

if __name__ == '__main__':
    pass
    # 训练集
    traindata = {
        "California Max": [3, 104, "爱情片"],
        "He's Not Really into Dudes": [2, 100, "爱情片"],
        "Beautiful": [1, 81, "爱情片"],
        "Kevin Longblande": [101, 10, "动作片"],
        "Robo Slayer": [99, 5, "动作片"],
        "Amped": [98, 2, "动作片"],
    }
    # 测试集
    testdata = {"Hi boy": [18, 90, "未知"]}

    # 提取数据
    tranlable = []  # 标签
    getList = []  # 训练集
    x1 = testdata["Hi boy"]
    x2 = testdata["Hi boy"]
    new_test = [x1[0], x2[1]]
    for i in traindata.keys():
        temp = traindata[i]
        getList.append(temp[0])  # x
        getList.append(temp[1])  # y
        tranlable.append(temp[2])  # 标签

    x = [i for i in getList[0::2]]
    y = [i for i in getList[1::2]]
    new_traindata = numpy.c_[x, y]

    # 测试集规格和训练集一致
    new_test = numpy.tile(new_test, (new_traindata.shape[0], 1))

    # 欧式距离
    distance = numpy.sqrt(numpy.sum((new_test - new_traindata) ** 2, axis=1))

    # 排序返回原下标
    sortList = numpy.argsort(distance)

    # 分类
    k = 3
    a = b = 0
    for i in sortList[0:k]:
        if tranlable[i] == "爱情片":
            a += 1
        else:
            b += 1
    print("爱情片") if a>b else print("动作片")

    # 数据可视化
    for i in range(new_traindata.shape[0]):
        if tranlable[i] == "爱情片":
            plt.scatter(new_traindata[i, 0], new_traindata[i, 1], c="r")
        else:
            plt.scatter(new_traindata[i, 0], new_traindata[i, 1], c="g")
    if a>b:
        plt.scatter(new_test[0, 0], new_test[0, 1], c="r", marker="+")
    else:
        plt.scatter(new_test[0, 0], new_test[0, 1], c="g", marker="+")
    plt.show()