def setUp(self):
self.hf_modules_cache = tempfile.mkdtemp()
self.cache_dir = tempfile.mkdtemp()
self.download_config = DownloadConfig(cache_dir=self.cache_dir)
self.dynamic_modules_path = datasets.load.init_dynamic_modules(
name="test_datasets_modules_" +
os.path.basename(self.hf_modules_cache),
hf_modules_cache=self.hf_modules_cache,
)
def test_extracted_datasets_path(default_extracted, default_cache_dir, xz_file, tmp_path, monkeypatch):
custom_cache_dir = "custom_cache"
custom_extracted_dir = "custom_extracted_dir"
custom_extracted_path = tmp_path / "custom_extracted_path"
if default_extracted:
expected = ("downloads" if default_cache_dir else custom_cache_dir, "extracted")
else:
monkeypatch.setattr("datasets.config.EXTRACTED_DATASETS_DIR", custom_extracted_dir)
monkeypatch.setattr("datasets.config.EXTRACTED_DATASETS_PATH", str(custom_extracted_path))
expected = custom_extracted_path.parts[-2:] if default_cache_dir else (custom_cache_dir, custom_extracted_dir)
filename = xz_file
download_config = (
DownloadConfig(extract_compressed_file=True)
if default_cache_dir
else DownloadConfig(cache_dir=tmp_path / custom_cache_dir, extract_compressed_file=True)
)
extracted_file_path = cached_path(filename, download_config=download_config)
assert Path(extracted_file_path).parent.parts[-2:] == expected
def test_cached_path_extract(compression_format, gz_file, xz_file, zstd_path, tmp_path, text_file):
input_paths = {"gzip": gz_file, "xz": xz_file, "zstd": zstd_path}
input_path = str(input_paths[compression_format])
cache_dir = tmp_path / "cache"
download_config = DownloadConfig(cache_dir=cache_dir, extract_compressed_file=True)
extracted_path = cached_path(input_path, download_config=download_config)
with open(extracted_path) as f:
extracted_file_content = f.read()
with open(text_file) as f:
expected_file_content = f.read()
assert extracted_file_content == expected_file_content
def load_builder_class(self, dataset_name, is_local=False):
# Download/copy dataset script
if is_local is True:
dataset_module = dataset_module_factory(
os.path.join("datasets", dataset_name))
else:
dataset_module = dataset_module_factory(
dataset_name,
download_config=DownloadConfig(force_download=True))
# Get dataset builder class
builder_cls = import_main_class(dataset_module.module_path)
return builder_cls
def test_dummy_data_autogenerate(self):
n_lines = 5
with TemporaryDirectory() as tmp_dir:
with open(os.path.join(tmp_dir, "train.txt"),
"w",
encoding="utf-8") as f:
f.write("foo\nbar\n" * 10)
with open(os.path.join(tmp_dir, "test.txt"), "w",
encoding="utf-8") as f:
f.write("foo\nbar\n" * 10)
class MockDownloadManagerWithCustomDatasetsScriptsDir(
MockDownloadManager):
datasets_scripts_dir = os.path.join(tmp_dir, "datasets")
cache_dir = os.path.join(tmp_dir, "cache")
os.makedirs(cache_dir, exist_ok=True)
dataset_builder = DummyBuilder(tmp_test_dir=tmp_dir,
cache_dir=cache_dir)
mock_dl_manager = MockDownloadManagerWithCustomDatasetsScriptsDir(
dataset_name=dataset_builder.name,
config=None,
version=Version("0.0.0"),
use_local_dummy_data=True,
cache_dir=cache_dir,
load_existing_dummy_data=False, # dummy data don't exist yet
)
download_config = DownloadConfig(cache_dir=os.path.join(
tmp_dir, datasets.config.DOWNLOADED_DATASETS_DIR))
dl_manager = DummyDataGeneratorDownloadManager(
dataset_name=dataset_builder.name,
mock_download_manager=mock_dl_manager,
download_config=download_config,
)
dataset_builder.download_and_prepare(dl_manager=dl_manager,
try_from_hf_gcs=False)
shutil.rmtree(dataset_builder._cache_dir)
dl_manager.auto_generate_dummy_data_folder(n_lines=n_lines)
path_do_dataset = os.path.join(
mock_dl_manager.datasets_scripts_dir,
mock_dl_manager.dataset_name)
dl_manager.compress_autogenerated_dummy_data(path_do_dataset)
mock_dl_manager.load_existing_dummy_data = True
dataset_builder.download_and_prepare(dl_manager=mock_dl_manager,
ignore_verifications=True,
try_from_hf_gcs=False)
dataset = dataset_builder.as_dataset(split="train")
self.assertEqual(len(dataset), n_lines)
del dataset
def test_load_real_dataset(self, dataset_name):
path = "./datasets/" + dataset_name
dataset_module = dataset_module_factory(
path, download_config=DownloadConfig(local_files_only=True))
builder_cls = import_main_class(dataset_module.module_path)
name = builder_cls.BUILDER_CONFIGS[
0].name if builder_cls.BUILDER_CONFIGS else None
with tempfile.TemporaryDirectory() as temp_cache_dir:
dataset = load_dataset(path,
name=name,
cache_dir=temp_cache_dir,
download_mode=DownloadMode.FORCE_REDOWNLOAD)
for split in dataset.keys():
self.assertTrue(len(dataset[split]) > 0)
del dataset
def test_download_manager_download(urls_type, tmp_path, monkeypatch):
import requests
monkeypatch.setattr(requests, "request", mock_request)
url = URL
if issubclass(urls_type, str):
urls = url
elif issubclass(urls_type, list):
urls = [url]
elif issubclass(urls_type, dict):
urls = {"train": url}
dataset_name = "dummy"
cache_subdir = "downloads"
cache_dir_root = str(tmp_path)
download_config = DownloadConfig(
cache_dir=os.path.join(cache_dir_root, cache_subdir),
use_etag=False,
)
dl_manager = DownloadManager(dataset_name=dataset_name, download_config=download_config)
downloaded_paths = dl_manager.download(urls)
input_urls = urls
for downloaded_paths in [downloaded_paths]:
if isinstance(urls, str):
downloaded_paths = [downloaded_paths]
input_urls = [urls]
elif isinstance(urls, dict):
assert "train" in downloaded_paths.keys()
downloaded_paths = downloaded_paths.values()
input_urls = urls.values()
assert downloaded_paths
for downloaded_path, input_url in zip(downloaded_paths, input_urls):
assert downloaded_path == dl_manager.downloaded_paths[input_url]
downloaded_path = Path(downloaded_path)
parts = downloaded_path.parts
assert parts[-1] == HASH
assert parts[-2] == cache_subdir
assert downloaded_path.exists()
content = downloaded_path.read_text()
assert content == CONTENT
metadata_downloaded_path = downloaded_path.with_suffix(".json")
assert metadata_downloaded_path.exists()
metadata_content = json.loads(metadata_downloaded_path.read_text())
assert metadata_content == {"url": URL, "etag": None}
def test_load_dataset_deletes_extracted_files(deleted, jsonl_gz_path,
tmp_path):
data_files = jsonl_gz_path
cache_dir = tmp_path / "cache"
if deleted:
download_config = DownloadConfig(delete_extracted=True,
cache_dir=cache_dir / "downloads")
ds = load_dataset("json",
split="train",
data_files=data_files,
cache_dir=cache_dir,
download_config=download_config)
else: # default
ds = load_dataset("json",
split="train",
data_files=data_files,
cache_dir=cache_dir)
assert ds[0] == {"col_1": "0", "col_2": 0, "col_3": 0.0}
assert (sorted(
(cache_dir / "downloads" / "extracted").iterdir()) == []) is deleted
def test_download_manager_extract(paths_type, xz_file, text_file):
filename = str(xz_file)
if issubclass(paths_type, str):
paths = filename
elif issubclass(paths_type, list):
paths = [filename]
elif issubclass(paths_type, dict):
paths = {"train": filename}
dataset_name = "dummy"
cache_dir = xz_file.parent
extracted_subdir = "extracted"
download_config = DownloadConfig(
cache_dir=cache_dir,
use_etag=False,
)
dl_manager = DownloadManager(dataset_name=dataset_name, download_config=download_config)
extracted_paths = dl_manager.extract(paths)
input_paths = paths
for extracted_paths in [extracted_paths]:
if isinstance(paths, str):
extracted_paths = [extracted_paths]
input_paths = [paths]
elif isinstance(paths, dict):
assert "train" in extracted_paths.keys()
extracted_paths = extracted_paths.values()
input_paths = paths.values()
assert extracted_paths
for extracted_path, input_path in zip(extracted_paths, input_paths):
assert extracted_path == dl_manager.extracted_paths[input_path]
extracted_path = Path(extracted_path)
parts = extracted_path.parts
assert parts[-1] == hash_url_to_filename(input_path, etag=None)
assert parts[-2] == extracted_subdir
assert extracted_path.exists()
extracted_file_content = extracted_path.read_text()
expected_file_content = text_file.read_text()
assert extracted_file_content == expected_file_content
def _autogenerate_dummy_data(self, dataset_builder, mock_dl_manager,
keep_uncompressed) -> Optional[bool]:
dl_cache_dir = (os.path.join(
self._cache_dir, config.DOWNLOADED_DATASETS_DIR) if self._cache_dir
else config.DOWNLOADED_DATASETS_PATH)
download_config = DownloadConfig(cache_dir=dl_cache_dir)
dl_manager = DummyDataGeneratorDownloadManager(
dataset_name=self._dataset_name,
mock_download_manager=mock_dl_manager,
download_config=download_config)
dataset_builder._split_generators(dl_manager)
mock_dl_manager.load_existing_dummy_data = False # don't use real dummy data
dl_manager.auto_generate_dummy_data_folder(
n_lines=self._n_lines,
json_field=self._json_field,
xml_tag=self._xml_tag,
match_text_files=self._match_text_files,
encoding=self._encoding,
)
if not keep_uncompressed:
path_do_dataset = os.path.join(
mock_dl_manager.datasets_scripts_dir,
mock_dl_manager.dataset_name)
dl_manager.compress_autogenerated_dummy_data(path_do_dataset)
# now test that the dummy_data.zip file actually works
mock_dl_manager.load_existing_dummy_data = True # use real dummy data
n_examples_per_split = {}
os.makedirs(dataset_builder._cache_dir, exist_ok=True)
try:
split_generators = dataset_builder._split_generators(
mock_dl_manager)
for split_generator in split_generators:
dataset_builder._prepare_split(split_generator,
check_duplicate_keys=False)
n_examples_per_split[
split_generator.
name] = split_generator.split_info.num_examples
except OSError as e:
logger.error(
f"Failed to load dummy data for config '{dataset_builder.config.name}''.\nOriginal error:\n"
+ str(e))
return False
else:
if all(n_examples > 0
for n_examples in n_examples_per_split.values()):
logger.warning(
f"Dummy data generation done and dummy data test succeeded for config '{dataset_builder.config.name}''."
)
return True
else:
empty_splits = [
split_name for split_name in n_examples_per_split
if n_examples_per_split[split_name] == 0
]
logger.warning(
f"Dummy data generation done but dummy data test failed since splits {empty_splits} have 0 examples for config '{dataset_builder.config.name}''."
)
return False
else:
generated_dummy_data_dir = os.path.join(
self._path_to_dataset, mock_dl_manager.dummy_data_folder)
logger.info(
f"Dummy data generated in directory '{generated_dummy_data_dir}' but kept uncompressed. "
"Please compress this directory into a zip file to use it for dummy data tests."
)
def run(self):
import apache_beam as beam
if self._name is not None and self._all_configs:
print(
"Both parameters `name` and `all_configs` can't be used at once."
)
exit(1)
path, name = self._dataset, self._name
dataset_module = dataset_module_factory(path)
builder_cls = import_main_class(dataset_module.module_path)
builders: List[DatasetBuilder] = []
if self._beam_pipeline_options:
beam_options = beam.options.pipeline_options.PipelineOptions(
flags=[
f"--{opt.strip()}"
for opt in self._beam_pipeline_options.split(",") if opt
])
else:
beam_options = None
if self._all_configs and len(builder_cls.BUILDER_CONFIGS) > 0:
for builder_config in builder_cls.BUILDER_CONFIGS:
builders.append(
builder_cls(
name=builder_config.name,
data_dir=self._data_dir,
hash=dataset_module.hash,
beam_options=beam_options,
cache_dir=self._cache_dir,
base_path=dataset_module.builder_kwargs.get(
"base_path"),
))
else:
builders.append(
builder_cls(
name=name,
data_dir=self._data_dir,
beam_options=beam_options,
cache_dir=self._cache_dir,
base_path=dataset_module.builder_kwargs.get("base_path"),
**self._config_kwargs,
))
for builder in builders:
builder.download_and_prepare(
download_mode=DownloadMode.REUSE_CACHE_IF_EXISTS if
not self._force_redownload else DownloadMode.FORCE_REDOWNLOAD,
download_config=DownloadConfig(
cache_dir=config.DOWNLOADED_DATASETS_PATH),
ignore_verifications=self._ignore_verifications,
try_from_hf_gcs=False,
)
if self._save_infos:
builder._save_infos()
print("Apache beam run successful.")
# If save_infos=True, the dataset infos file is created next to the loaded module file.
# Let's move it to the original directory of the dataset script, to allow the user to
# upload them on S3 at the same time afterwards.
if self._save_infos:
dataset_infos_path = os.path.join(
builder_cls.get_imported_module_dir(),
config.DATASETDICT_INFOS_FILENAME)
name = Path(path).name + ".py"
combined_path = os.path.join(path, name)
if os.path.isfile(path):
dataset_dir = os.path.dirname(path)
elif os.path.isfile(combined_path):
dataset_dir = path
else: # in case of a remote dataset
print(f"Dataset Infos file saved at {dataset_infos_path}")
exit(1)
# Move datasetinfo back to the user
user_dataset_infos_path = os.path.join(
dataset_dir, config.DATASETDICT_INFOS_FILENAME)
copyfile(dataset_infos_path, user_dataset_infos_path)
print(f"Dataset Infos file saved at {user_dataset_infos_path}")