torch.utils.data.DataLoader与迭代器转换操作
在做实验时,我们常常会使用用开源的数据集进行测试。而Pytorch中内置了许多数据集,这些数据集我们常常使用DataLoader
类进行加载。
如下面这个我们使用DataLoader
类加载torch.vision
中的FashionMNIST
数据集。
from torch.utils.data import DataLoader from torchvision import datasets from torchvision.transforms import ToTensor import matplotlib.pyplot as plt training_data = datasets.FashionMNIST( root="data", train=True, download=True, transform=ToTensor() ) test_data = datasets.FashionMNIST( root="data", train=False, download=True, transform=ToTensor() )
我们接下来定义Dataloader对象用于加载这两个数据集:
train_dataloader = DataLoader(training_data, batch_size=64, shuffle=True) test_dataloader = DataLoader(test_data, batch_size=64, shuffle=True)
那么这个train_dataloader
究竟是什么类型呢?
print(type(train_dataloader)) # <class 'torch.utils.data.dataloader.DataLoader'>
我们可以将先其转换为迭代器类型。
print(type(iter(train_dataloader)))# <class 'torch.utils.data.dataloader._SingleProcessDataLoaderIter'>
然后再使用next(iter(train_dataloader))
从迭代器里取数据,如下所示:
train_features, train_labels = next(iter(train_dataloader)) print(f"Feature batch shape: {train_features.size()}") print(f"Labels batch shape: {train_labels.size()}") img = train_features[0].squeeze() label = train_labels[0] plt.imshow(img, cmap="gray") plt.show() print(f"Label: {label}")
可以看到我们成功获取了数据集中第一张图片的信息,控制台打印:
Feature batch shape: torch.Size([64, 1, 28, 28]) Labels batch shape: torch.Size([64]) Label: 2
图片可视化显示如下:
不过有读者可能就会产生疑问,很多时候我们并没有将DataLoader类型强制转换成迭代器类型呀,大多数时候我们会写如下代码:
for train_features, train_labels in train_dataloader: print(train_features.shape) # torch.Size([64, 1, 28, 28]) print(train_features[0].shape) # torch.Size([1, 28, 28]) print(train_features[0].squeeze().shape) # torch.Size([28, 28]) img = train_features[0].squeeze() label = train_labels[0] plt.imshow(img, cmap="gray") plt.show() print(f"Label: {label}")
可以看到,该代码也能够正常迭代训练数据,前三个样本的控制台打印输出为:
torch.Size([64, 1, 28, 28]) torch.Size([1, 28, 28]) torch.Size([28, 28]) Label: 7 torch.Size([64, 1, 28, 28]) torch.Size([1, 28, 28]) torch.Size([28, 28]) Label: 4 torch.Size([64, 1, 28, 28]) torch.Size([1, 28, 28]) torch.Size([28, 28]) Label: 1
那么为什么我们这里没有显式将Dataloader
转换为迭代器类型呢,其实是Python语言for循环的一种机制,一旦我们用for ... in ...句式来迭代一个对象,那么Python
解释器就会偷偷地自动帮我们创建好迭代器,也就是说
for train_features, train_labels in train_dataloader:
实际上等同于
for train_features, train_labels in iter(train_dataloader):
更进一步,这实际上等同于
train_iterator = iter(train_dataloader) try: while True: train_features, train_labels = next(train_iterator) except StopIteration: pass
推而广之,我们在用Python迭代直接迭代列表时:
for x in [1, 2, 3, 4]:
其实Python解释器已经为我们隐式转换为迭代器了:
list_iterator = iter([1, 2, 3, 4]) try: while True: x = next(list_iterator) except StopIteration: pass
栏 目:Python代码
本文标题:torch.utils.data.DataLoader与迭代器转换操作
本文地址:http://www.codeinn.net/misctech/222171.html