Skip to content

GitLab

Unverified Commit e946e87d authored by Philip Meier's avatar Philip Meier Committed by GitHub
Browse files

refactor download tests (#7546)

parent 6381f7b2
Hide whitespace changes
Inline Side-by-side
Showing with 132 additions and 242 deletions
test/test_datasets_download.py
View file @ e946e87d
...@@ -14,13 +14,7 @@ from urllib.request import Request, urlopen ...@@ -14,13 +14,7 @@ from urllib.request import Request, urlopen
import pytest import pytest
from torchvision import datasets from torchvision import datasets
from torchvision.datasets.utils import ( from torchvision.datasets.utils import _get_redirect_url, USER_AGENT
_get_redirect_url,
check_integrity,
download_file_from_google_drive,
download_url,
USER_AGENT,
)
def limit_requests_per_time(min_secs_between_requests=2.0): def limit_requests_per_time(min_secs_between_requests=2.0):
...@@ -84,47 +78,45 @@ urlopen = resolve_redirects()(urlopen) ...@@ -84,47 +78,45 @@ urlopen = resolve_redirects()(urlopen)
@contextlib.contextmanager @contextlib.contextmanager
def log_download_attempts( def log_download_attempts(
urls_and_md5s=None, urls,
file="utils", *,
patch=True, dataset_module,
mock_auxiliaries=None,
): ):
def add_mock(stack, name, file, **kwargs): def maybe_add_mock(*, module, name, stack, lst=None):
patcher = unittest.mock.patch(f"torchvision.datasets.{module}.{name}")
try: try:
return stack.enter_context(unittest.mock.patch(f"torchvision.datasets.{file}.{name}", **kwargs)) mock = stack.enter_context(patcher)
except AttributeError as error: except AttributeError:
if file != "utils": return
return add_mock(stack, name, "utils", **kwargs)
else:
raise pytest.UsageError from error
if urls_and_md5s is None:
urls_and_md5s = set()
if mock_auxiliaries is None:
mock_auxiliaries = patch
with contextlib.ExitStack() as stack: if lst is not None:
url_mock = add_mock(stack, "download_url", file, wraps=None if patch else download_url) lst.append(mock)
google_drive_mock = add_mock(
stack, "download_file_from_google_drive", file, wraps=None if patch else download_file_from_google_drive
)
if mock_auxiliaries: with contextlib.ExitStack() as stack:
add_mock(stack, "extract_archive", file) download_url_mocks = []
download_file_from_google_drive_mocks = []
for module in [dataset_module, "utils"]:
maybe_add_mock(module=module, name="download_url", stack=stack, lst=download_url_mocks)
maybe_add_mock(
module=module,
name="download_file_from_google_drive",
stack=stack,
lst=download_file_from_google_drive_mocks,
)
maybe_add_mock(module=module, name="extract_archive", stack=stack)
try: try:
yield urls_and_md5s yield
finally: finally:
for args, kwargs in url_mock.call_args_list: for download_url_mock in download_url_mocks:
url = args[0] for args, kwargs in download_url_mock.call_args_list:
md5 = args[-1] if len(args) == 4 else kwargs.get("md5") urls.append(args[0] if args else kwargs["url"])
urls_and_md5s.add((url, md5))
for args, kwargs in google_drive_mock.call_args_list: for download_file_from_google_drive_mock in download_file_from_google_drive_mocks:
id = args[0] for args, kwargs in download_file_from_google_drive_mock.call_args_list:
url = f"https://drive.google.com/file/d/{id}" file_id = args[0] if args else kwargs["file_id"]
md5 = args[3] if len(args) == 4 else kwargs.get("md5") urls.append(f"https://drive.google.com/file/d/{file_id}")
urls_and_md5s.add((url, md5))
def retry(fn, times=1, wait=5.0): def retry(fn, times=1, wait=5.0):
...@@ -170,45 +162,14 @@ def assert_url_is_accessible(url, timeout=5.0): ...@@ -170,45 +162,14 @@ def assert_url_is_accessible(url, timeout=5.0):
urlopen(request, timeout=timeout) urlopen(request, timeout=timeout)
def assert_file_downloads_correctly(url, md5, tmpdir, timeout=5.0): def collect_urls(dataset_cls, *args, **kwargs):
file = path.join(tmpdir, path.basename(url)) urls = []
with assert_server_response_ok(): with contextlib.suppress(Exception), log_download_attempts(
with open(file, "wb") as fh: urls, dataset_module=dataset_cls.__module__.split(".")[-1]
request = Request(url, headers={"User-Agent": USER_AGENT}) ):
response = urlopen(request, timeout=timeout) dataset_cls(*args, **kwargs)
fh.write(response.read())
assert check_integrity(file, md5=md5), "The MD5 checksums mismatch"
class DownloadConfig:
def __init__(self, url, md5=None, id=None):
self.url = url
self.md5 = md5
self.id = id or url
def __repr__(self) -> str:
return self.id
def make_download_configs(urls_and_md5s, name=None): return [(url, f"{dataset_cls.__name__}, {url}") for url in urls]
return [
DownloadConfig(url, md5=md5, id=f"{name}, {url}" if name is not None else None) for url, md5 in urls_and_md5s
]
def collect_download_configs(dataset_loader, name=None, **kwargs):
urls_and_md5s = set()
try:
with log_download_attempts(urls_and_md5s=urls_and_md5s, **kwargs):
dataset = dataset_loader()
except Exception:
dataset = None
if name is None and dataset is not None:
name = type(dataset).__name__
return make_download_configs(urls_and_md5s, name)
# This is a workaround since fixtures, such as the built-in tmp_dir, can only be used within a test but not within a # This is a workaround since fixtures, such as the built-in tmp_dir, can only be used within a test but not within a
...@@ -223,12 +184,14 @@ def root(): ...@@ -223,12 +184,14 @@ def root():
def places365(): def places365():
return itertools.chain( return itertools.chain.from_iterable(
*[ [
collect_download_configs( collect_urls(
lambda: datasets.Places365(ROOT, split=split, small=small, download=True), datasets.Places365,
name=f"Places365, {split}, {'small' if small else 'large'}", ROOT,
file="places365", split=split,
small=small,
download=True,
) )
for split, small in itertools.product(("train-standard", "train-challenge", "val"), (False, True)) for split, small in itertools.product(("train-standard", "train-challenge", "val"), (False, True))
] ]
...@@ -236,30 +199,26 @@ def places365(): ...@@ -236,30 +199,26 @@ def places365():
def caltech101(): def caltech101():
return collect_download_configs(lambda: datasets.Caltech101(ROOT, download=True), name="Caltech101") return collect_urls(datasets.Caltech101, ROOT, download=True)
def caltech256(): def caltech256():
return collect_download_configs(lambda: datasets.Caltech256(ROOT, download=True), name="Caltech256") return collect_urls(datasets.Caltech256, ROOT, download=True)
def cifar10(): def cifar10():
return collect_download_configs(lambda: datasets.CIFAR10(ROOT, download=True), name="CIFAR10") return collect_urls(datasets.CIFAR10, ROOT, download=True)
def cifar100(): def cifar100():
return collect_download_configs(lambda: datasets.CIFAR100(ROOT, download=True), name="CIFAR100") return collect_urls(datasets.CIFAR100, ROOT, download=True)
def voc(): def voc():
# TODO: Also test the "2007-test" key # TODO: Also test the "2007-test" key
return itertools.chain( return itertools.chain.from_iterable(
*[ [
collect_download_configs( collect_urls(datasets.VOCSegmentation, ROOT, year=year, download=True)
lambda: datasets.VOCSegmentation(ROOT, year=year, download=True),
name=f"VOC, {year}",
file="voc",
)
for year in ("2007", "2008", "2009", "2010", "2011", "2012") for year in ("2007", "2008", "2009", "2010", "2011", "2012")
] ]
) )
...@@ -267,59 +226,42 @@ def voc(): ...@@ -267,59 +226,42 @@ def voc():
def mnist(): def mnist():
with unittest.mock.patch.object(datasets.MNIST, "mirrors", datasets.MNIST.mirrors[-1:]): with unittest.mock.patch.object(datasets.MNIST, "mirrors", datasets.MNIST.mirrors[-1:]):
return collect_download_configs(lambda: datasets.MNIST(ROOT, download=True), name="MNIST") return collect_urls(datasets.MNIST, ROOT, download=True)
def fashion_mnist(): def fashion_mnist():
return collect_download_configs(lambda: datasets.FashionMNIST(ROOT, download=True), name="FashionMNIST") return collect_urls(datasets.FashionMNIST, ROOT, download=True)
def kmnist(): def kmnist():
return collect_download_configs(lambda: datasets.KMNIST(ROOT, download=True), name="KMNIST") return collect_urls(datasets.KMNIST, ROOT, download=True)
def emnist(): def emnist():
# the 'split' argument can be any valid one, since everything is downloaded anyway # the 'split' argument can be any valid one, since everything is downloaded anyway
return collect_download_configs(lambda: datasets.EMNIST(ROOT, split="byclass", download=True), name="EMNIST") return collect_urls(datasets.EMNIST, ROOT, split="byclass", download=True)
def qmnist(): def qmnist():
return itertools.chain( return itertools.chain.from_iterable(
*[ [collect_urls(datasets.QMNIST, ROOT, what=what, download=True) for what in ("train", "test", "nist")]
collect_download_configs(
lambda: datasets.QMNIST(ROOT, what=what, download=True),
name=f"QMNIST, {what}",
file="mnist",
)
for what in ("train", "test", "nist")
]
) )
def moving_mnist(): def moving_mnist():
return collect_download_configs(lambda: datasets.MovingMNIST(ROOT, download=True), name="MovingMNIST") return collect_urls(datasets.MovingMNIST, ROOT, download=True)
def omniglot(): def omniglot():
return itertools.chain( return itertools.chain.from_iterable(
*[ [collect_urls(datasets.Omniglot, ROOT, background=background, download=True) for background in (True, False)]
collect_download_configs(
lambda: datasets.Omniglot(ROOT, background=background, download=True),
name=f"Omniglot, {'background' if background else 'evaluation'}",
)
for background in (True, False)
]
) )
def phototour(): def phototour():
return itertools.chain( return itertools.chain.from_iterable(
*[ [
collect_download_configs( collect_urls(datasets.PhotoTour, ROOT, name=name, download=True)
lambda: datasets.PhotoTour(ROOT, name=name, download=True),
name=f"PhotoTour, {name}",
file="phototour",
)
# The names postfixed with '_harris' point to the domain 'matthewalunbrown.com'. For some reason all # The names postfixed with '_harris' point to the domain 'matthewalunbrown.com'. For some reason all
# requests timeout from within CI. They are disabled until this is resolved. # requests timeout from within CI. They are disabled until this is resolved.
for name in ("notredame", "yosemite", "liberty") # "notredame_harris", "yosemite_harris", "liberty_harris" for name in ("notredame", "yosemite", "liberty") # "notredame_harris", "yosemite_harris", "liberty_harris"
...@@ -328,91 +270,51 @@ def phototour(): ...@@ -328,91 +270,51 @@ def phototour():
def sbdataset(): def sbdataset():
return collect_download_configs( return collect_urls(datasets.SBDataset, ROOT, download=True)
lambda: datasets.SBDataset(ROOT, download=True),
name="SBDataset",
file="voc",
)
def sbu(): def sbu():
return collect_download_configs( return collect_urls(datasets.SBU, ROOT, download=True)
lambda: datasets.SBU(ROOT, download=True),
name="SBU",
file="sbu",
)
def semeion(): def semeion():
return collect_download_configs( return collect_urls(datasets.SEMEION, ROOT, download=True)
lambda: datasets.SEMEION(ROOT, download=True),
name="SEMEION",
file="semeion",
)
def stl10(): def stl10():
return collect_download_configs( return collect_urls(datasets.STL10, ROOT, download=True)
lambda: datasets.STL10(ROOT, download=True),
name="STL10",
)
def svhn(): def svhn():
return itertools.chain( return itertools.chain.from_iterable(
*[ [collect_urls(datasets.SVHN, ROOT, split=split, download=True) for split in ("train", "test", "extra")]
collect_download_configs(
lambda: datasets.SVHN(ROOT, split=split, download=True),
name=f"SVHN, {split}",
file="svhn",
)
for split in ("train", "test", "extra")
]
) )
def usps(): def usps():
return itertools.chain( return itertools.chain.from_iterable(
*[ [collect_urls(datasets.USPS, ROOT, train=train, download=True) for train in (True, False)]
collect_download_configs(
lambda: datasets.USPS(ROOT, train=train, download=True),
name=f"USPS, {'train' if train else 'test'}",
file="usps",
)
for train in (True, False)
]
) )
def celeba(): def celeba():
return collect_download_configs( return collect_urls(datasets.CelebA, ROOT, download=True)
lambda: datasets.CelebA(ROOT, download=True),
name="CelebA",
file="celeba",
)
def widerface(): def widerface():
return collect_download_configs( return collect_urls(datasets.WIDERFace, ROOT, download=True)
lambda: datasets.WIDERFace(ROOT, download=True),
name="WIDERFace",
file="widerface",
)
def kinetics(): def kinetics():
return itertools.chain( return itertools.chain.from_iterable(
*[ [
collect_download_configs( collect_urls(
lambda: datasets.Kinetics( datasets.Kinetics,
path.join(ROOT, f"Kinetics{num_classes}"), path.join(ROOT, f"Kinetics{num_classes}"),
frames_per_clip=1, frames_per_clip=1,
num_classes=num_classes, num_classes=num_classes,
split=split, split=split,
download=True, download=True,
),
name=f"Kinetics, {num_classes}, {split}",
file="kinetics",
) )
for num_classes, split in itertools.product(("400", "600", "700"), ("train", "val")) for num_classes, split in itertools.product(("400", "600", "700"), ("train", "val"))
] ]
...@@ -420,58 +322,55 @@ def kinetics(): ...@@ -420,58 +322,55 @@ def kinetics():
def kitti(): def kitti():
return itertools.chain( return itertools.chain.from_iterable(
*[ [collect_urls(datasets.Kitti, ROOT, train=train, download=True) for train in (True, False)]
collect_download_configs(
lambda train=train: datasets.Kitti(ROOT, train=train, download=True),
name=f"Kitti, {'train' if train else 'test'}",
file="kitti",
)
for train in (True, False)
]
) )
def make_parametrize_kwargs(download_configs): def stanford_cars():
argvalues = [] return itertools.chain.from_iterable(
ids = [] [collect_urls(datasets.StanfordCars, ROOT, split=split, download=True) for split in ["train", "test"]]
for config in download_configs: )
argvalues.append((config.url, config.md5))
ids.append(config.id)
def url_parametrization(*dataset_urls_and_ids_fns):
return dict(argnames=("url", "md5"), argvalues=argvalues, ids=ids) return pytest.mark.parametrize(
"url",
[
@pytest.mark.parametrize( pytest.param(url, id=id)
**make_parametrize_kwargs( for dataset_urls_and_ids_fn in dataset_urls_and_ids_fns
itertools.chain( for url, id in sorted(set(dataset_urls_and_ids_fn()))
caltech101(), ],
caltech256(),
cifar10(),
cifar100(),
# The VOC download server is unstable. See https://github.com/pytorch/vision/issues/2953 for details.
# voc(),
mnist(),
fashion_mnist(),
kmnist(),
emnist(),
qmnist(),
omniglot(),
phototour(),
sbdataset(),
semeion(),
stl10(),
svhn(),
usps(),
celeba(),
widerface(),
kinetics(),
kitti(),
places365(),
)
) )
@url_parametrization(
caltech101,
caltech256,
cifar10,
cifar100,
# The VOC download server is unstable. See https://github.com/pytorch/vision/issues/2953 for details.
# voc,
mnist,
fashion_mnist,
kmnist,
emnist,
qmnist,
omniglot,
phototour,
sbdataset,
semeion,
stl10,
svhn,
usps,
celeba,
widerface,
kinetics,
kitti,
places365,
sbu,
) )
def test_url_is_accessible(url, md5): def test_url_is_accessible(url):
""" """
If you see this test failing, find the offending dataset in the parametrization and move it to If you see this test failing, find the offending dataset in the parametrization and move it to
``test_url_is_not_accessible`` and link an issue detailing the problem. ``test_url_is_not_accessible`` and link an issue detailing the problem.
...@@ -479,15 +378,11 @@ def test_url_is_accessible(url, md5): ...@@ -479,15 +378,11 @@ def test_url_is_accessible(url, md5):
retry(lambda: assert_url_is_accessible(url)) retry(lambda: assert_url_is_accessible(url))
@pytest.mark.parametrize( @url_parametrization(
**make_parametrize_kwargs( stanford_cars, # https://github.com/pytorch/vision/issues/7545
itertools.chain(
sbu(), # https://github.com/pytorch/vision/issues/7005
)
)
) )
@pytest.mark.xfail @pytest.mark.xfail
def test_url_is_not_accessible(url, md5): def test_url_is_not_accessible(url):
""" """
As the name implies, this test is the 'inverse' of ``test_url_is_accessible``. Since the download servers are As the name implies, this test is the 'inverse' of ``test_url_is_accessible``. Since the download servers are
beyond our control, some files might not be accessible for longer stretches of time. Still, we want to know if they beyond our control, some files might not be accessible for longer stretches of time. Still, we want to know if they
...@@ -497,8 +392,3 @@ def test_url_is_not_accessible(url, md5): ...@@ -497,8 +392,3 @@ def test_url_is_not_accessible(url, md5):
``test_url_is_accessible``. ``test_url_is_accessible``.
""" """
retry(lambda: assert_url_is_accessible(url)) retry(lambda: assert_url_is_accessible(url))
@pytest.mark.parametrize(**make_parametrize_kwargs(itertools.chain()))
def test_file_downloads_correctly(url, md5):
retry(lambda: assert_file_downloads_correctly(url, md5))
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment