numpy KNN简单实现

numpy支持矩阵减向量（每行减去某个向量） 列的情况可以考虑转置。

R实现：
library(class)
attach(iris)
train <- iris[,1:2]
y <- as.numeric(Species)
x <- train
fit <- knn(x, x, y)
print(1 - sum(fit==y)/length(y))

利用ndarray的诸多特性，可以实现非常简短的（DIY）KNN：

import numpy as np 
from sklearn.datasets import load_iris
iris = load_iris()
#print iris.DESCR
X = iris.data[:,1:3]
y = iris.target 

def predict_y(x, k = 1):
 distance_array = np.linalg.norm(X - x, axis = 1)
 # the increase distance indexs
 distance_indexs = np.argsort(distance_array)[1:k + 1]
 y_unique_counts = np.unique(y[distance_indexs], return_counts = True)

 return y_unique_counts[0][np.argmax(y_unique_counts[1])]

predict = map(predict_y, X)
error_num = 0
for i in range(len(predict)):
 if predict[i] != y[i]:
  error_num += 1

print "error_error :"
print float(error_num) / len(predict)