def prepare_loaders_impl(
location: str,
train_ratio: float = 0.9,
vote_ratio: float = 0.05,
batch_size: int = 32,
dp_enabled: bool = False,
) -> Tuple[PrefetchDataset, PrefetchDataset, PrefetchDataset]:
data_folder = get_data(location)
images = pickle.load(open(Path(data_folder) / IMAGE_FL, "rb"))
labels = pickle.load(open(Path(data_folder) / LABEL_FL, "rb"))
n_cases = int(train_ratio * len(images))
n_vote_cases = int(vote_ratio * len(images))
train_loader = _make_loader(images[:n_cases],
labels[:n_cases],
batch_size,
dp_enabled=dp_enabled)
vote_loader = _make_loader(images[n_cases:n_cases + n_vote_cases],
labels[n_cases:n_cases + n_vote_cases],
batch_size)
test_loader = _make_loader(images[n_cases + n_vote_cases:],
labels[n_cases + n_vote_cases:], batch_size)
return train_loader, vote_loader, test_loader
def prepare_data_loaders(
location: str,
train_ratio: float = 0.9,
vote_ratio: float = 0.05,
batch_size: int = 32,
) -> Tuple[PrefetchDataset, PrefetchDataset, PrefetchDataset]:
"""
Load training data from folders and create train and test dataloader
:param location: Path to training dataset
:param train_ratio: What portion of train_data should be used as test set
:param batch_size:
:return: Tuple of train_loader and test_loader
"""
data_folder = get_data(location)
images = pickle.load(open(Path(data_folder) / IMAGE_FL, "rb"))
labels = pickle.load(open(Path(data_folder) / LABEL_FL, "rb"))
n_cases = int(train_ratio * len(images))
n_vote_cases = int(vote_ratio * len(images))
train_loader = _make_loader(images[:n_cases], labels[:n_cases], batch_size)
vote_loader = _make_loader(images[n_cases:n_cases + n_vote_cases],
labels[n_cases:n_cases + n_vote_cases],
batch_size)
test_loader = _make_loader(images[n_cases + n_vote_cases:],
labels[n_cases + n_vote_cases:], batch_size)
return train_loader, vote_loader, test_loader
def prepare_data_loaders(
location: str,
train_ratio: float = 0.8,
vote_ratio: float = 0.1,
) -> Tuple[Tuple[np.ndarray, np.ndarray], Tuple[np.ndarray, np.ndarray], Tuple[
np.ndarray, np.ndarray]]:
"""
Load training data from folders and create train and test arrays
:param location: Path to training dataset
:param train_ratio: What portion of train_data should be used as test set
:return: Tuple of tuples (train_data, train_labels), (test_data, test_loaders)
"""
data_folder = get_data(location)
data = pickle.load(open(Path(data_folder) / DATA_FL, "rb"))
labels = pickle.load(open(Path(data_folder) / LABEL_FL, "rb"))
n_cases = int(train_ratio * len(data))
n_vote_cases = int(vote_ratio * len(data))
assert (n_cases > 0), "There are no cases"
train_data, train_labels = data[:n_cases], labels[:n_cases]
vote_data, vote_labels = data[n_cases:n_cases +
n_vote_cases], labels[n_cases:n_cases +
n_vote_cases]
test_data, test_labels = data[n_cases +
n_vote_cases:], labels[n_cases +
n_vote_cases:]
return (train_data, train_labels), (vote_data, vote_labels), (test_data,
test_labels)
def prepare_data_loaders(
location: str,
train_ratio: float = 0.8,
vote_ratio: float = 0.1,
batch_size: int = 8,
no_cuda: bool = False,
) -> Tuple[DataLoader, DataLoader, DataLoader]:
"""
Load training data from folders and create train and test dataloader
:param location: Path to training dataset
:param train_ratio: What portion of train_data should be used as test set
:param batch_size: Batch size
:param no_cuda: Disable GPU computing
:return: Tuple of train_loader and test_loader
"""
cuda = not no_cuda and torch.cuda.is_available()
DataloaderKwargs = TypedDict('DataloaderKwargs', {
'num_workers': int,
'pin_memory': bool
},
total=False)
loader_kwargs: DataloaderKwargs = {
'num_workers': 1,
'pin_memory': True
} if cuda else {}
data_folder = get_data(location)
data = pickle.load(open(Path(data_folder) / DATA_FL, "rb"))
labels = pickle.load(open(Path(data_folder) / LABEL_FL, "rb"))
n_cases = int(train_ratio * len(data))
n_vote_cases = int(vote_ratio * len(data))
assert (n_cases > 0), "There are no cases"
train_loader = _make_loader(data[:n_cases], labels[:n_cases], batch_size,
**loader_kwargs)
vote_loader = _make_loader(data[n_cases:n_cases + n_vote_cases],
labels[n_cases:n_cases + n_vote_cases],
batch_size)
test_loader = _make_loader(data[n_cases + n_vote_cases:],
labels[n_cases + n_vote_cases:], batch_size,
**loader_kwargs)
return train_loader, vote_loader, test_loader
def prepare_data_loaders(train_folder: str,
train_ratio: float = 0.9,
batch_size: int = 32,
**_kwargs) -> Tuple[PrefetchDataset, PrefetchDataset]:
"""
Load training data from folders and create train and test dataloader
:param train_folder: Path to training dataset
:param train_ratio: What portion of train_data should be used as test set
:param batch_size:
:param _kwargs: Residual parameters not used by this function
:return: Tuple of train_loader and test_loader
"""
data_folder = get_data(train_folder)
images = pickle.load(open(Path(data_folder) / IMAGE_FL, "rb"))
labels = pickle.load(open(Path(data_folder) / LABEL_FL, "rb"))
n_cases = int(train_ratio * len(images))
train_loader = _make_loader(images[:n_cases], labels[:n_cases], batch_size)
test_loader = _make_loader(images[n_cases:], labels[n_cases:], batch_size)
return train_loader, test_loader
def prepare_data_loaders(
location: str,
test_location: Optional[str] = None,
train_ratio: float = 0.96,
batch_size: int = 8,
no_cuda: bool = False,
) -> Tuple[DataLoader, DataLoader]:
"""
Load training data from folders and create train and test dataloader
:param location: Path to training dataset
:param test_location: Path to test dataset
:param train_ratio: When test_location is not specified what portion of train_data should be used as test set
:param batch_size:
:param no_cuda: Disable GPU computing
:return: Tuple of train_loader and test_loader
"""
cuda = not no_cuda and torch.cuda.is_available()
DataloaderKwargs = TypedDict('DataloaderKwargs', {
'num_workers': int,
'pin_memory': bool
},
total=False)
loader_kwargs: DataloaderKwargs = {
'num_workers': 1,
'pin_memory': True
} if cuda else {}
data_folder = get_data(location)
if test_location is not None:
local_test_folder = get_data(test_location)
train_loader = DataLoader(XrayDataset(data_folder,
train=True,
train_ratio=1.0),
batch_size=batch_size,
shuffle=True,
**loader_kwargs)
test_loader = DataLoader(XrayDataset(local_test_folder,
train=True,
train_ratio=1.0),
batch_size=batch_size,
shuffle=True,
**loader_kwargs)
else:
train_loader = DataLoader(XrayDataset(data_folder,
train=True,
train_ratio=train_ratio),
batch_size=batch_size,
shuffle=True,
**loader_kwargs)
test_loader = DataLoader(XrayDataset(data_folder,
train=False,
train_ratio=train_ratio),
batch_size=batch_size,
shuffle=True,
**loader_kwargs)
return train_loader, test_loader
