一、线性扫描法

  1. from tensorflow import keras
  2. import numpy as np
  3. import logging
  5. logging.basicConfig(level=logging.INFO,format='%(asctime)s - %(name)s - %(levelname)s - %(message)s')
  6. logger = logging.getLogger(__name__)
  8. def calDist(x1,x2):
  9.     '''
  10.     计算两个点之间的距离
  11.     '''
  12.     # 欧氏距离
  13.     return np.sqrt(np.sum(np.square(x1-x2)))
  14.     # 等同于求差值向量的l2范数
  15.     # return np.linalg.norm(x1-x2)
  17. def getClostest(trainData,trainLabel,x,topK):
  18.     '''
  19.     预测样本x的类别
  20.     '''
  21.     # distList = np.zeros(trainLabel.shape[0])
  22.     distList = [0] * len(trainLabel)
  25.     for i in range(len(trainData)):
  26.         xi = trainData[i]
  27.         iDist = calDist(xi,x)
  28.         distList[i] = iDist
  30.     # 返回前k近的样本的序号
  31.     topKList = np.argsort(distList)[:topK]
  33.     labelList = [0] *10
  34.     for index in topKList:
  35.         labelList[int(trainLabel[index])]+=1
  36.     result = labelList.index(max(labelList))
  37.     return result
  39. if __name__ == "__main__":
  40.     (x_train, y_train), (x_test, y_test)=keras.datasets.mnist.load_data()
  42.     errCnt = 0
  43.     for i in range(100):
  44.         x = x_test[i]
  45.         y = getClostest(x_train,y_train,x,25)
  46.         logger.info((y,y_test[i]))
  47.         if y!=y_test[i]: errCnt+=1    
  48.     logger.info('correct rate:%d%%' % (100-errCnt))

二、kd树