knn----依靠一个点的基础实现

knn–依靠一个点的基础实现

knn是一个基础的分类的算法，它不像其他的神经网络算法，他只是通过简单的比较某个维度的几何距离（？？？）来实现分类的

from sklearn.datasets import load_iris
import matplotlib.pyplot as plt 
from mpl_toolkits.mplot3d import Axes3D
import random

iris_datasets = load_iris()

trainData = iris_datasets['data'][:]

data0 = [i for j in trainData for i in j[:1]]
data1 = [i for j in trainData for i in j[1:2]]
data2 = [i for j in trainData for i in j[2:3]]
data3 = [i for j in trainData for i in j[3:4]]


kind0 = [i for i in range(50)]
kind1 = [i for i in range(50,100)]
kind2 = [i for i in range(100,150)]

########################
ax = plt.subplot(111,projection = '3d')
plt.title("First")
ax.scatter(data0[:50],data1[:50],data2[:50],c = 'r')
ax.scatter(data0[50:100],data1[50:100],data2[50:100],c = 'g')
ax.scatter(data0[100:150],data1[100:150],data2[100:150],c = 'b')
# ax,scatter()
ax.set_xlabel("X")
ax.set_ylabel("Y")
ax.set_zlabel("Z")
plt.show()
#plt.close()
########################
plt.title("Second")
bx = plt.subplot(111,projection = '3d')
bx.scatter(data3[:50],data1[:50],data2[:50],c = 'r')
bx.scatter(data3[50:100],data1[50:100],data2[50:100],c = 'g')
bx.scatter(data3[100:150],data1[100:150],data2[100:150],c = 'b')
# ax,scatter()
bx.set_xlabel("X")
bx.set_ylabel("Y")
bx.set_zlabel("Z")
plt.show()
#plt.close()
###########################
plt.title("Third")
cx = plt.subplot(111,projection = '3d')
cx.scatter(data3[:50],data0[:50],data2[:50],c = 'r')
cx.scatter(data3[50:100],data0[50:100],data2[50:100],c = 'g')
cx.scatter(data3[100:150],data0[100:150],data2[100:150],c = 'b')
# ax,scatter()
cx.set_xlabel("X")
cx.set_ylabel("Y")
cx.set_zlabel("Z")
plt.show()
plt.close()
#########################


testData = []
realValue = []
for _ in range(5):
    a = random.randint(0,149)
    realValue.append(a//50)
    flag = 0
    if a in trainData:
        flag = 1
    if a in kind0 and flag == 1:
        del kind0[a]
    elif a in kind1 and flag == 1:
        del kind1[a-50]
    else:
        if flag == 1:
            del kind2[a-100]

    testData.append([data0[a],data1[a],data2[a],data3[a]])
    if flag != 1:
        del data0[a]
        del data1[a]
        del data2[a]
        del data3[a]

# print(testData,'/',trainData)

def distence(array1 = [0,0,0,0], array2 = [0,0,0,0]):
    if array1.__len__() != array2.__len__():
        print(r"error")
        return
    num = 0
    for _ in range(array1.__len__()):
        num  = num + (array1[_] - array2[_])**2
    
    return num


def predict(array1 = [0,0,0,0]):
    if array1.__len__() != 4:
        print(r"error")
        return -1
    variety = distence(array1,[data0[0],data1[0],data2[0],data3[0]])
    flag = 0

    for _ in range(1,145):
        if(variety>distence(array1,[data0[_],data1[_],data2[_],data3[_]])):
            variety = distence(array1,[data0[_],data1[_],data2[_],data3[_]])
            flag = _
    # print(flag)
    return flag//50

per = 0
for i in range(testData.__len__()):
    print("预测值",predict(testData[i]),"真实值",realValue[i])
    if(predict(testData[i]) == realValue[i]):
        per = per + 1
    if(i == testData.__len__()-1):
        print("识别成功率：",per*100/testData.__len__(),"%\n")

这里只是一个点的简单实现，多个点的实现思路也一样，只是需要将多个点的集合距离相加来比较处最值

使用matplotlib来可视化数据的分类，但是由于具有四个参数，无法绘制出四维图形，因此也可以通过绘制四张三维的图来表示。