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_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=GenerateMode.FORCE_REDOWNLOAD)
for split in dataset.keys():
self.assertTrue(len(dataset[split]) > 0)
del dataset
def run(self):
set_verbosity_warning()
dataset_module = dataset_module_factory(self._path_to_dataset)
builder_cls = import_main_class(dataset_module.module_path)
# use `None` as config if no configs
builder_configs = builder_cls.BUILDER_CONFIGS or [None]
auto_generate_results = []
with tempfile.TemporaryDirectory() as tmp_dir:
for builder_config in builder_configs:
if builder_config is None:
name = None
version = builder_cls.VERSION
else:
version = builder_config.version
name = builder_config.name
dataset_builder = builder_cls(name=name,
hash=dataset_module.hash,
cache_dir=tmp_dir)
mock_dl_manager = MockDownloadManager(
dataset_name=self._dataset_name,
config=builder_config,
version=version,
use_local_dummy_data=True,
load_existing_dummy_data=False,
)
if self._auto_generate:
auto_generate_results.append(
self._autogenerate_dummy_data(
dataset_builder=dataset_builder,
mock_dl_manager=mock_dl_manager,
keep_uncompressed=self._keep_uncompressed,
))
else:
self._print_dummy_data_instructions(
dataset_builder=dataset_builder,
mock_dl_manager=mock_dl_manager)
if self._auto_generate and not self._keep_uncompressed:
if all(auto_generate_results):
print(
f"Automatic dummy data generation succeeded for all configs of '{self._path_to_dataset}'"
)
else:
print(
f"Automatic dummy data generation failed for some configs of '{self._path_to_dataset}'"
)
def test_load_real_dataset_all_configs(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)
config_names = ([
config.name for config in builder_cls.BUILDER_CONFIGS
] if len(builder_cls.BUILDER_CONFIGS) > 0 else [None])
for name in config_names:
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_dataset_info_available(self, dataset, config_name):
with TemporaryDirectory() as tmp_dir:
dataset_module = dataset_module_factory(os.path.join(
"datasets", dataset),
cache_dir=tmp_dir,
local_files_only=True)
builder_cls = import_main_class(dataset_module.module_path,
dataset=True)
builder_instance: DatasetBuilder = builder_cls(
cache_dir=tmp_dir,
name=config_name,
hash=dataset_module.hash,
)
dataset_info_url = os.path.join(
HF_GCP_BASE_URL,
builder_instance._relative_data_dir(with_hash=False),
config.DATASET_INFO_FILENAME).replace(os.sep, "/")
datset_info_path = cached_path(dataset_info_url, cache_dir=tmp_dir)
self.assertTrue(os.path.exists(datset_info_path))
def run(self):
fl_logger().setLevel(ERROR)
if self._name is not None and self._all_configs:
print(
"Both parameters `config` and `all_configs` can't be used at once."
)
exit(1)
path, name = self._dataset, self._name
module = dataset_module_factory(path)
builder_cls = import_main_class(module.module_path)
if self._all_configs and len(builder_cls.BUILDER_CONFIGS) > 0:
n_builders = len(builder_cls.BUILDER_CONFIGS) // self._num_proc
n_builders += (len(builder_cls.BUILDER_CONFIGS) %
self._num_proc) > self._proc_rank
else:
n_builders = 1 if self._proc_rank == 0 else 0
def get_builders() -> Generator[DatasetBuilder, None, None]:
if self._all_configs and len(builder_cls.BUILDER_CONFIGS) > 0:
for i, config in enumerate(builder_cls.BUILDER_CONFIGS):
if i % self._num_proc == self._proc_rank:
if "name" in module.builder_kwargs:
yield builder_cls(
cache_dir=self._cache_dir,
data_dir=self._data_dir,
**module.builder_kwargs,
)
else:
yield builder_cls(
name=config.name,
cache_dir=self._cache_dir,
data_dir=self._data_dir,
**module.builder_kwargs,
)
else:
if self._proc_rank == 0:
if "name" in module.builder_kwargs:
yield builder_cls(cache_dir=self._cache_dir,
data_dir=self._data_dir,
**module.builder_kwargs)
else:
yield builder_cls(name=name,
cache_dir=self._cache_dir,
data_dir=self._data_dir,
**module.builder_kwargs)
for j, builder in enumerate(get_builders()):
print(
f"Testing builder '{builder.config.name}' ({j + 1}/{n_builders})"
)
builder._record_infos = True
builder.download_and_prepare(
download_mode=DownloadMode.REUSE_CACHE_IF_EXISTS if
not self._force_redownload else DownloadMode.FORCE_REDOWNLOAD,
ignore_verifications=self._ignore_verifications,
try_from_hf_gcs=False,
)
builder.as_dataset()
if self._save_infos:
builder._save_infos()
# 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(),
datasets.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
elif os.path.isdir(
path
): # for local directories containing only data files
dataset_dir = path
else: # in case of a remote dataset
dataset_dir = None
print(f"Dataset Infos file saved at {dataset_infos_path}")
# Move dataset_info back to the user
if dataset_dir is not None:
user_dataset_infos_path = os.path.join(
dataset_dir,
datasets.config.DATASETDICT_INFOS_FILENAME)
copyfile(dataset_infos_path, user_dataset_infos_path)
print(
f"Dataset Infos file saved at {user_dataset_infos_path}"
)
# If clear_cache=True, the download folder and the dataset builder cache directory are deleted
if self._clear_cache:
if os.path.isdir(builder._cache_dir):
logger.warning(f"Clearing cache at {builder._cache_dir}")
rmtree(builder._cache_dir)
download_dir = os.path.join(
self._cache_dir, datasets.config.DOWNLOADED_DATASETS_DIR)
if os.path.isdir(download_dir):
logger.warning(f"Clearing cache at {download_dir}")
rmtree(download_dir)
print("Test successful.")
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"),
namespace=dataset_module.builder_kwargs.get(
"namespace"),
))
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"),
namespace=dataset_module.builder_kwargs.get("namespace"),
))
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),
save_infos=self._save_infos,
ignore_verifications=self._ignore_verifications,
try_from_hf_gcs=False,
)
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}")
