KNN 分类、回归底层

import numpy as np
from scipy.stats import mode
from sklearn.datasets import load_boston

T = np.array([[3, 104, -1],
    [2, 100, -1],
    [1, 81, -1],
    [101, 10, 1],
    [99, 5, 1],
    [98, 2, 1]])
x = np.array([18, 90])

trainX=T[:,:-1]
trainY=T[:,-1]

普通knn分类

def knn_class(trainX,trainY,testX,k=5):
    dist=np.sqrt(np.sum(np.square(trainX-testX),axis=1))
    select=trainY[np.argsort(dist)][:k]
    return mode(select)[0][0]

knn_class(trainX,trainY,x)

-1

加权knn分类

def knn_weight_class(trainX,trainY,testX,k=5):
    weight=np.sqrt(np.sum(np.square(trainX-testX),axis=1))**-1
    sort_arg=np.argsort(weight)[-k:]
    weight_sum=[weight[sort_arg][trainY[sort_arg]==i].sum() for i in np.unique(trainY)]
    return np.unique(trainY)[np.argmax(weight_sum)]

knn_weight_class(trainX,trainY,x)

-1

boston=load_boston()
x=boston.data
y=boston.target

trainX=x[:-1]
trainY=y[:-1]
testX=x[-1]
testY=y[-1]

普通knn回归

def knn_regression(trainX,trainY,testX,k=5):
    dist=np.sqrt(np.sum(np.square(trainX-testX),axis=1))
    select=trainY[np.argsort(dist)][:k]
    return np.mean(select)

pred=knn_regression(trainX,trainY,testX)
print(pred,testY)

23.759999999999998 11.9

加权knn回归

def knn_weight_regression(trainX,trainY,testX,k=5):
    weight=np.sqrt(np.sum(np.square(trainX-testX),axis=1))**-1
    sort_arg=np.argsort(weight)[-k:]

    a=weight[sort_arg].sum()**-1
    res=np.sum(trainY[sort_arg]*weight[sort_arg])

    return res*a

pred=knn_weight_regression(trainX,trainY,testX)
print(pred,testY)

23.112713355853334 11.9