def _prepare_data(self):
params = self.params
train_set = mdl.get_dataset("MNIST", split="train")
valid_set = mdl.get_dataset("MNIST", split="valid")
def get_loader(dataset):
l = torch.utils.data.DataLoader(
dataset,
batch_size=param.batch_size,
drop_last=True,
shuffle=True,
num_workers=4,
)
return l
train_loader = get_loader(train_set)
valid_loader = get_loader(valid_set)
params.len_train_batch = len(train_loader)
params.len_test_batch = len(train_loader)
iters = {
"train": ELoaderIter(train_loader),
"valid": ELoaderIter(valid_loader),
}
return None, iters
def _prepare_data(self):
'''
prepare your dataset here
and return a iterator dic
'''
params = self.params
train_set = mdl.get_dataset("MNIST", split="train")
valid_set = mdl.get_dataset("MNIST", split="valid")
def get_loader(dataset):
l = torch.utils.data.DataLoader(
dataset,
batch_size=param.batch_size,
drop_last=True,
shuffle=True,
num_workers=4,
)
return l
train_loader = get_loader(train_set)
valid_loader = get_loader(valid_set)
iters = {
"train": ELoaderIter(train_loader),
"valid": ELoaderIter(valid_loader),
}
return None, iters
def get_set(dataset, domain, split):
if dataset is None or dataset == "NONE":
dataset = mdl.get_dataset(
dataset=domain, domain=None, split=split
)
else:
dataset = mdl.get_dataset(
dataset=dataset, domain=domain, split=split
)
return dataset
def _prepare_data(self):
params = self.params
dataset = params.dataset
source = params.source
target = params.target
def get_set(dataset, domain, split):
if dataset is None or dataset == "NONE":
dataset = mdl.get_dataset(
dataset=domain, domain=None, split=split
)
else:
dataset = mdl.get_dataset(
dataset=dataset, domain=domain, split=split
)
return dataset
train_S_set = mdl.get_dataset(dataset, source, split="train")
train_T_set = mdl.get_dataset(dataset, target, split="train")
valid_set = mdl.get_dataset(dataset, target, split="test")
def get_loader(dataset, shuffle, drop_last, batch_size=None):
batch_size = (
params.batch_size if batch_size is None else batch_size
)
l = torch.utils.data.DataLoader(
dataset,
batch_size=params.batch_size,
drop_last=drop_last,
shuffle=shuffle,
)
return l
train_S_l = get_loader(train_S_set, shuffle=True, drop_last=True)
train_T_l = get_loader(train_T_set, shuffle=True, drop_last=True)
valid_l = get_loader(
valid_set,
shuffle=True,
drop_last=True,
batch_size=params.batch_size / 2,
)
iters = {
"train": {
"S": ELoaderIter(train_S_l),
"T": ELoaderIter(train_T_l),
},
"valid": ELoaderIter(valid_l),
}
return None, iters