目录
- 正文
- 1. 加载数据集
- 2. 迭代和可视化数据集
- 3.创建自定义数据集
- 3.1 __init__
- 3.2 __len__
- 3.3 __getitem__
- 4. 使用 DataLoaders 为训练准备数据
- 5.遍历 DataLoader
正文
处理数据样本的代码可能会逐渐变得混乱且难以维护;理想情况下,我们希望我们的数据集代码与我们的模型训练代码分离,以获得更好的可读性和模块化。PyTorch 提供了两个数据原语:torch.utils.data.DataLoader
和torch.utils.data.Dataset
允许我们使用预加载的数据集以及自定义数据。 Dataset
存储样本及其对应的标签,DataLoader
封装了一个迭代器用于遍历Dataset
,以便轻松访问样本数据。
PyTorch 领域库提供了许多预加载的数据集(例如 FashionMNIST),这些数据集继承自torch.utils.data.Dataset
并实现了特定于特定数据的功能。它们可用于对您的模型进行原型设计和基准测试。你可以在这里找到它们:图像数据集、 文本数据集和 音频数据集
1. 加载数据集
下面是如何从 TorchVision 加载Fashion-MNIST数据集的示例。Fashion-MNIST 是 Zalando 文章图像的数据集,由 60,000 个训练示例和 10,000 个测试示例组成。每个示例都包含 28×28 灰度图像和来自 10 个类别之一的相关标签。
我们使用以下参数加载FashionMNIST 数据集:
root
是存储训练/测试数据的路径,train
指定训练或测试数据集,download=True
如果数据不可用,则从 Internet 下载数据root
。transform
并target_transform
指定特征和标签转换
import torch
from torch.utils.data import Dataset
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()
)
Downloading http://fashion-mnist.s3-website.eu-central-1.amazonaws.com/train-images-idx3-ubyte.gz
Downloading http://fashion-mnist.s3-website.eu-central-1.amazonaws.com/train-images-idx3-ubyte.gz to data/FashionMNIST/raw/train-images-idx3-ubyte.gz
0%| | 0/26421880 [00:00<?, ?it/s]
0%| | 32768/26421880 [00:00<01:26, 303914.51it/s]
0%| | 65536/26421880 [00:00<01:27, 301769.74it/s]
0%| | 131072/26421880 [00:00<01:00, 437795.76it/s]
1%| | 229376/26421880 [00:00<00:42, 621347.43it/s]
2%|1 | 491520/26421880 [00:00<00:20, 1259673.64it/s]
4%|3 | 950272/26421880 [00:00<00:11, 2264911.11it/s]
7%|7 | 1933312/26421880 [00:00<00:05, 4467299.81it/s]
15%|#4 | 3833856/26421880 [00:00<00:02, 8587616.55it/s]
26%|##6 | 6881280/26421880 [00:00<00:01, 14633777.99it/s]
37%|###7 | 9830400/26421880 [00:01<00:00, 18150145.01it/s]
49%|####8 | 12910592/26421880 [00:01<00:00, 21161097.17it/s]
61%|###### | 16023552/26421880 [00:01<00:00, 23366004.89it/s]
72%|#######2 | 19136512/26421880 [00:01<00:00, 24967488.10it/s]
84%|########4 | 22249472/26421880 [00:01<00:00, 26016258.24it/s]
95%|#########5| 25231360/26421880 [00:01<00:00, 26218488.24it/s]
100%|##########| 26421880/26421880 [00:01<00:00, 15984902.80it/s]
Extracting data/FashionMNIST/raw/train-images-idx3-ubyte.gz to data/FashionMNIST/raw
Downloading http://fashion-mnist.s3-website.eu-central-1.amazonaws.com/train-labels-idx1-ubyte.gz
Downloading http://fashion-mnist.s3-website.eu-central-1.amazonaws.com/train-labels-idx1-ubyte.gz to data/FashionMNIST/raw/train-labels-idx1-ubyte.gz
0%| | 0/29515 [00:00<?, ?it/s]
100%|##########| 29515/29515 [00:00<00:00, 268356.24it/s]
100%|##########| 29515/29515 [00:00<00:00, 266767.69it/s]
Extracting data/FashionMNIST/raw/train-labels-idx1-ubyte.gz to data/FashionMNIST/raw
Downloading http://fashion-mnist.s3-website.eu-central-1.amazonaws.com/t10k-images-idx3-ubyte.gz
Downloading http://fashion-mnist.s3-website.eu-central-1.amazonaws.com/t10k-images-idx3-ubyte.gz to data/FashionMNIST/raw/t10k-images-idx3-ubyte.gz
0%| | 0/4422102 [00:00<?, ?it/s]
1%| | 32768/4422102 [00:00<00:14, 302027.13it/s]
1%|1 | 65536/4422102 [00:00<00:14, 300501.69it/s]
3%|2 | 131072/4422102 [00:00<00:09, 436941.45it/s]
5%|5 | 229376/4422102 [00:00<00:06, 619517.19it/s]
10%|9 | 425984/4422102 [00:00<00:03, 1044158.55it/s]
20%|## | 884736/4422102 [00:00<00:01, 2114396.73it/s]
40%|#### | 1769472/4422102 [00:00<00:00, 4067080.68it/s]
80%|######## | 3538944/4422102 [00:00<00:00, 7919346.09it/s]
100%|##########| 4422102/4422102 [00:00<00:00, 5036535.17it/s]
Extracting data/FashionMNIST/raw/t10k-images-idx3-ubyte.gz to data/FashionMNIST/raw
Downloading http://fashion-mnist.s3-website.eu-central-1.amazonaws.com/t10k-labels-idx1-ubyte.gz
Downloading http://fashion-mnist.s3-website.eu-central-1.amazonaws.com/t10k-labels-idx1-ubyte.gz to data/FashionMNIST/raw/t10k-labels-idx1-ubyte.gz
0%| | 0/5148 [00:00<?, ?it/s]
100%|##########| 5148/5148 [00:00<00:00, 22168662.21it/s]
Extracting data/FashionMNIST/raw/t10k-labels-idx1-ubyte.gz to data/FashionMNIST/raw
2. 迭代和可视化数据集
我们可以像python 列表一样索引Datasets
,比如:
training_data[index]
.
我们用matplotlib
来可视化训练数据中的一些样本。
labels_map = {
0: "T-Shirt",
1: "Trouser",
2: "Pullover",
3: "Dress",
4: "Coat",
5: "Sandal",
6: "Shirt",
7: "Sneaker",
8: "Bag",
9: "Ankle Boot",
}
figure = plt.figure(figsize=(8, 8))
cols, rows = 3, 3
for i in range(1, cols * rows + 1):
sample_idx = torch.randint(len(training_data), size=(1,)).item()
img, label = training_data[sample_idx]
figure.add_subplot(rows, cols, i)
plt.title(labels_map[label])
plt.axis("off")
plt.imshow(img.squeeze(), cmap="gray")
plt.show()
3.创建自定义数据集
自定义 Dataset 类必须实现三个函数:init、len__和__getitem。
比如: FashionMNIST 图像存储在一个目录img_dir
中,它们的标签分别存储在一个 CSV 文件annotations_file
中。
在接下来的部分中,我们将分析每个函数中发生的事情。
import os
import pandas as pd
from torchvision.io import read_image
class CustomImageDataset(Dataset):
def __init__(self, annotations_file, img_dir, transform=None, target_transform=None):
self.img_labels = pd.read_csv(annotations_file)
self.img_dir = img_dir
self.transform = transform
self.target_transform = target_transform
def __len__(self):
return len(self.img_labels)
def __getitem__(self, idx):
img_path = os.path.join(self.img_dir, self.img_labels.iloc[idx, 0])
image = read_image(img_path)
label = self.img_labels.iloc[idx, 1]
if self.transform:
image = self.transform(image)
if self.target_transform:
label = self.target_transform(label)
return image, label
3.1 __init__
init 函数在实例化 Dataset 对象时运行一次。我们初始化包含图像、注释文件和两种转换的目录(在下一节中更详细地介绍)。
labels.csv 文件如下所示:
tshirt1.jpg, 0
tshirt2.jpg, 0
......
ankleboot999.jpg, 9
def __init__(self, annotations_file, img_dir, transform=None, target_transform=None):
self.img_labels = pd.read_csv(annotations_file)
self.img_dir = img_dir
self.transform = transform
self.target_transform = target_transform
3.2 __len__
len 函数返回我们数据集中的样本数。
例子:
def __len__(self):
return len(self.img_labels)
3.3 __getitem__
getitem 函数从给定索引处的数据集中加载并返回一个样本idx
。基于索引,它识别图像在磁盘上的位置,使用 将其转换为张量read_image
,从 csv 数据中检索相应的标签self.img_labels
,调用它们的转换函数(如果适用),并返回张量图像和相应的标签一个元组。
def __getitem__(self, idx):
img_path = os.path.join(self.img_dir, self.img_labels.iloc[idx, 0])
image = read_image(img_path)
label = self.img_labels.iloc[idx, 1]
if self.transform:
image = self.transform(image)
if self.target_transform:
label = self.target_transform(label)
return image, label
4. 使用 DataLoaders 为训练准备数据
Dataset
一次加载一个样本数据和其对应的label。在训练模型时,我们通常希望以minibatches“小批量”的形式传递样本,在每个 epoch 重新洗牌以减少模型过拟合,并使用 Pythonmultiprocessing
加速数据检索。
DataLoader
是一个可迭代对象,它封装了复杂性并暴漏了简单的API。
from torch.utils.data import DataLoader
train_dataloader = DataLoader(training_data, batch_size=64, shuffle=True)
test_dataloader = DataLoader(test_data, batch_size=64, shuffle=True)
5.遍历 DataLoader
我们已将该数据集加载到 DataLoader
中,并且可以根据需要遍历数据集。下面的每次迭代都会返回一批train_features
和train_labels
(分别包含batch_size=64
特征和标签)。因为我们指定shuffle=True
了 ,所以在我们遍历所有批次之后,数据被打乱(为了更细粒度地控制数据加载顺序,请查看Samplers)。
# Display image and label.
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}")
Feature batch shape: torch.Size([64, 1, 28, 28])
Labels batch shape: torch.Size([64])
Label: 4