实现了一个小demo演示SubsetRandomSampler的用法
import torch
from torch.utils.data import TensorDataset, DataLoader
from torch.utils.data.sampler import SubsetRandomSampler
dataset = TensorDataset(torch.tensor(list(range(20)))) # 构造一个数据集（0到19）
idx = list(range(len(dataset))) # 创建索引，SubsetRandomSampler会自动乱序
# idx = torch.zeros(len(dataset)).long() # 传入相同的索引，SubsetRandomSampler只会采样相同结果
n = len(dataset)
split = n//5
train_sampler = SubsetRandomSampler(idx[split::]) # 随机取80%的数据做训练集
test_sampler = SubsetRandomSampler(idx[::split]) # 随机取20%的数据做测试集
train_loader = DataLoader(dataset, sampler=train_sampler)
test_loader = DataLoader(dataset, sampler=test_sampler)
print('data for training:')
for i in train_loader:
print(i)
print('data for testing:')
for i in test_loader:
print(i)
注意train_loader和test_loader的dataset都是一样的，比如要获取loader的样本总数，应该len(sampler)而不是len(dataset)
len(train_loader.sampler)
len(test_loader.sampler)
关于pytorch的其他sampler的文档：
https://blog.csdn.net/aiwanghuan5017/article/details/102147825
