KNN实现Cifar10数据的分类.zip

import numpy as np import torch from torch.utils.data import DataLoader from torch.utils.data import Dataset import torchvision.datasets as dsets import matplotlib.pyplot as plt import operator import warnings warnings.filterwarnings('ignore') batch_size = 100 # Cifar10 dataset train_dataset = dsets.CIFAR10(root='/ml/pycifar', # 选择数据的根目录 train=True, # 选择训练集 download=True) # 从网上下载图片 test_dataset = dsets.CIFAR10(root='/ml/pycifar', # 选择数据的根目录 train=False, # 选择测试集 download=True) # 从网上下载图片 #加载数据集 train_loader = torch.utils.data.DataLoader(dataset=train_dataset,batch_size=batch_size,shuffle=True) test_loader = torch.utils.data.DataLoader(dataset=test_dataset ,batch_size=batch_size,shuffle=True) classes = ('plane','car','bird','cat','deer','dog','flog','horse','ship','truck') digit = train_loader.dataset.data[20000] # plt.imshow(digit,cmap=plt.cm.binary) # plt.show() # print(classes[train_loader.dataset.targets[20000]]) def getmean(X_train): X_train = np.reshape(X_train,(X_train.shape[0],-1)) # 将图片从二维展开为一维 mean_image = np.mean(X_train,axis=0) # 求解出训练集中所有图片每个像素位置上的平均值 return mean_image def centralized(X_test,mean_image): X_test = np.reshape(X_test,(X_test.shape[0],-1)) # 将图片从二维展开为一维 X_test = X_test.astype(np.float) # 将数据类型转为浮点型 X_test = X_test - mean_image # 减去均值图像，实现零均值化 return X_test def KNN_classify(k,dis,X_train,Y_train,X_test): # assert dis == 'E' or dis == 'M' 'dis must E or M ; E代表欧式距离，M代表曼哈顿距离' num_test = X_test.shape[0] #表示测试数据集的个数 labellist = [] ''' 使用欧式距离公式作为距离度量 ''' if (dis == 'E'): for i in range(num_test): #实现欧式距离的计算公式 distances = np.sqrt(np.sum(((X_train - np.tile(X_test[i],(X_train.shape[0],1))) ** 2) , axis=1)) nearest_k = np.argsort(distances) #距离由小到大进行排序，并返回 index 值 topk = nearest_k[:k] # 选取前 K 个距离的点 classCount = {} for j in topk: # 依次遍历这 K 个点 classCount[Y_train[j]] = classCount.get(Y_train[j],0) + 1 # operator.itemgetter(1) 表示取出对象维度为 1 的数据 sortedClassCount = sorted(classCount.items(),key=operator.itemgetter(1),reverse=True) labellist.append(sortedClassCount[0][0]) return np.array(labellist) ''' 使用曼哈顿距离公式作为距离度量 ''' if (dis == 'M'): for i in range(num_test): #实现曼哈顿距离的计算公式 distances = np.sum(np.abs(X_train - np.tile(X_test[i],(X_train.shape[0],1))),axis=1) nearset_k = np.argsort(distances) #距离由大到小进行排序，并返回 index 值 topk = nearset_k[:k] # 选取前 K 个距离的点 ''' 下面开始进行类别计数，并对类别号进行排序，输出类别数目最多的作为该点的类别 ''' classCount = {} # 定义了一个字典 for j in topk: # 依次遍历这 K 个点 # 注意此处的 get 函数，因为没有设置 Y_train[j] 这个键的值，因此默认输出为 0 classCount[Y_train[j]] = classCount.get(Y_train[j],0) + 1 #类别计算器，用来计算类别 sortedClassCount = sorted(classCount.items(),key=operator.itemgetter(1),reverse=True) #对类别计数器进行排序，选择类别最多的一个作为该测试集的类别 labellist.append(sortedClassCount[0][0]) return np.array(labellist) X_train = train_loader.dataset.data mean_image1 = getmean(X_train) X_train = centralized(X_train,mean_image1) y_train = train_loader.dataset.targets X_test = test_loader.dataset.data[:100] mean_image2 = getmean(X_test) X_test = centralized(X_test,mean_image2) y_test = test_loader.dataset.targets[:100] num_test = len(y_test) y_test_pred = KNN_classify(6,'E',X_train,y_train,X_test) #使用没有封装好的类 num_correct = np.sum(y_test_pred == y_test) accuracy = float(num_correct)/num_test print('KNN实现Cifar10数据分类得精确度为'+str(accuracy*100)+'%')