需要在训练集里划出一部分作为验证集,可以使用SubsetRandomSampler
(或者其他的sampler
),例如:
from torch.utils.data.sampler import SubsetRandomSampler
#训练集样本数
num_training_samples = 48000
train_sampler = SubsetRandomSampler(torch.arange(0, num_training_samples))
#验证集样本数
num_val_samples = 12000
val_sampler = SubsetRandomSampler(torch.arange(num_training_samples, num_training_samples+num_val_samples))
然后在DataLoader
接口中如下设置即可,注意shuffle
此处必须为False
:
train_dataloader = torch.utils.data.DataLoader(
...
sampler=train_sampler,
...
)
val_dataloader = torch.utils.data.DataLoader(
...
sampler=val_sampler,
...
)
或者,也可以这也划分:
train_db, val_db = torch.utils.data.random_split(data, [48000, 12000])
train_dataloader = t.utils.data.DataLoader(
train_db,
...
)
# val_dataloader = t.utils.data.DataLoader(
val_db
...
)