def _load_client(
team_slug: Optional[str] = None,
offline: bool = False,
maybe_guest: bool = False,
dataset_identifier: Optional[str] = None,
):
"""Fetches a client, potentially offline
Parameters
----------
offline : bool
Flag for using an offline client
maybe_guest : bool
Flag to make a guest client, if config is missing
Returns
-------
Client
The client requested
"""
if not team_slug and dataset_identifier:
team_slug = DatasetIdentifier.parse(dataset_identifier).team_slug
try:
config_dir = Path.home() / ".darwin" / "config.yaml"
client = Client.from_config(config_dir, team_slug=team_slug)
return client
except MissingConfig:
if maybe_guest:
return Client.from_guest()
else:
_error("Authenticate first")
except InvalidLogin:
_error("Please re-authenticate")
except Unauthenticated:
_error("Please re-authenticate")
def it_returns_remote_files(darwin_client: Client):
dataset_id = 1
endpoint: str = f"/datasets/{dataset_id}/items?page%5Bsize%5D=500&page%5Bfrom%5D=0"
responses.add(responses.POST,
darwin_client.url + endpoint,
json={},
status=200)
darwin_client.fetch_remote_files(dataset_id, {
"page[size]": 500,
"page[from]": 0
}, {}, "v7")
def run_demo(
*,
team_slug: Optional[str],
dataset_slug: Optional[str] = None,
datasets_dir: Optional[str] = None,
api_key: Optional[str] = None,
config_path: Optional[Path] = None,
):
"""
Download a Darwin dataset on the file system.
Parameters
----------
team_slug : str
Slug of the team to select
dataset_slug : str
This is the dataset name with everything lower-case, removed specials characters and
spaces are replaced by dashes, e.g., `bird-species`. This string is unique within a team
datasets_dir : Path
Path where the client should be initialized from (aka the root path)
api_key: str
API key to authenticate the client
config_path: Path
Path to a configuration path which contains the authentication information to use
Returns
-------
splits : dict
Keys are the different splits (random, tags, ...) and values are the relative file names
"""
# Authenticate the new KEY if available
if api_key is not None:
authenticate(api_key=api_key,
default_team=True,
datasets_dir=datasets_dir)
# Get the client used to perform remote operations
if config_path is not None:
client = Client.from_config(config_path=config_path)
else:
client = Client.local(team_slug=team_slug)
# Create a dataset identifier
dataset_identifier = DatasetIdentifier.from_slug(dataset_slug=dataset_slug,
team_slug=team_slug)
# Get an object representing the remote dataset
ds = client.get_remote_dataset(dataset_identifier=dataset_identifier)
# Download the dataset on the local file system
ds.pull()
# Split the dataset in train/val/test
splits = split_dataset(dataset=ds)
def it_raises_if_workflow_id_is_not_found(darwin_client: Client):
item_id: int = 1234
endpoint: str = f"/dataset_items/{item_id}/workflow"
json_response: Dict[str, Any] = {}
responses.add(responses.POST,
darwin_client.url + endpoint,
json=json_response,
status=200)
with pytest.raises(ValueError) as exception:
darwin_client.instantitate_item(item_id)
assert str(
exception.value) == f"No Workflow Id found for item_id: {item_id}"
def darwin_client(darwin_config_path: Path, darwin_datasets_path: Path, team_slug: str) -> Client:
config = Config(darwin_config_path)
config.put(["global", "api_endpoint"], "http://localhost/api")
config.put(["global", "base_url"], "http://localhost")
config.put(["teams", team_slug, "api_key"], "mock_api_key")
config.put(["teams", team_slug, "datasets_dir"], str(darwin_datasets_path))
return Client(config=config)
def it_raises_if_comment_id_is_not_found(darwin_client: Client):
workflow_id = 1234
endpoint: str = f"/workflows/{workflow_id}/workflow_comment_threads"
json_response: Dict[str, Any] = {}
responses.add(responses.POST,
darwin_client.url + endpoint,
json=json_response,
status=200)
with pytest.raises(ValueError) as exception:
darwin_client.post_workflow_comment(workflow_id, "My comment.")
assert str(
exception.value
) == f"Unable to retrieve comment id for workflow: {workflow_id}."
def remote_dataset(dataset_slug: str, local_config_file: Config):
client = Client(local_config_file)
return RemoteDataset(client=client,
team="v7",
name="TEST_DATASET",
slug=dataset_slug,
dataset_id=1)
def it_returns_the_created_dataset(darwin_client: Client):
endpoint: str = "/datasets"
json_response: Dict[str, Any] = {
"name": "my-dataset",
"slug": "my-dataset",
"id": 1,
"num_images": 1,
"num_videos": 0,
"progress": 0,
}
responses.add(responses.POST,
darwin_client.url + endpoint,
json=json_response,
status=200)
actual_dataset = darwin_client.create_dataset("my-dataset", "v7")
expected_dataset = RemoteDataset(
team="v7",
name="my-dataset",
slug="my-dataset",
dataset_id=1,
item_count=1,
client=darwin_client,
)
assert_dataset(actual_dataset, expected_dataset)
def it_returns_the_dataset(darwin_client: Client):
endpoint: str = "/datasets"
json_response = [{
"name": "dataset-name-1",
"slug": "dataset-slug-1",
"id": 1,
"num_images": 1,
"num_videos": 0,
"progress": 0,
}]
responses.add(responses.GET,
darwin_client.url + endpoint,
json=json_response,
status=200)
actual_dataset = darwin_client.get_remote_dataset("v7/dataset-slug-1")
expected_dataset = RemoteDataset(
team="v7",
name="dataset-name-1",
slug="dataset-slug-1",
dataset_id=1,
item_count=1,
client=darwin_client,
)
assert_dataset(actual_dataset, expected_dataset)
def get_dataset_files(dataset_slug) -> Iterator[DatasetItem]:
""" Return a list of filenames in a dataset along with their status """
client = Client.from_api_key(API_KEY)
identifier = DatasetIdentifier.parse(dataset_slug)
dataset = client.get_remote_dataset(dataset_identifier=identifier)
resp = dataset.fetch_remote_files()
return resp
def add_labels_to_dataset(dataset_slug: str, labels: List[str], label_type: str):
""" Add labels to a dataset """
assert label_type in ['polygon', 'tag']
client = Client.from_api_key(API_KEY)
identifier = DatasetIdentifier.parse(dataset_slug)
dataset = client.get_remote_dataset(dataset_identifier=identifier)
for label in labels:
dataset.create_annotation_class(label, label_type)
def it_raises_if_dataset_is_not_found(darwin_client: Client):
endpoint: str = "/datasets"
json_response = [{
"name": "dataset-name-1",
"slug": "dataset-slug-1",
"id": 1,
"num_images": 1,
"num_videos": 0,
"progress": 0,
}]
responses.add(responses.GET,
darwin_client.url + endpoint,
json=json_response,
status=200)
with pytest.raises(NotFound):
darwin_client.get_remote_dataset("v7/dataset-slug-2")
def populate_dataset_annotations(dataset_slug, format_name: str, file_paths: List[str]):
assert format_name in ['darwin', 'coco', 'pascal_voc']
client = Client.from_api_key(API_KEY)
identifier = DatasetIdentifier.parse(dataset_slug)
dataset = client.get_remote_dataset(dataset_identifier=identifier)
format_dict = {k: v for (k, v) in darwin.importer.formats.supported_formats}
parser = format_dict[format_name]
importer.import_annotations(dataset, parser, file_paths)
def it_returns_workflow_id(darwin_client: Client):
item_id: int = 1234
workflow_id: int = 1
endpoint: str = f"/dataset_items/{item_id}/workflow"
json_response: Dict[str, Any] = {"current_workflow_id": workflow_id}
responses.add(responses.POST,
darwin_client.url + endpoint,
json=json_response,
status=200)
assert darwin_client.instantitate_item(item_id) == workflow_id
def it_returns_comment_id(darwin_client: Client):
comment_id: int = 1234
workflow_id: int = 1
endpoint: str = f"/workflows/{workflow_id}/workflow_comment_threads"
json_response: Dict[str, Any] = {"id": comment_id}
responses.add(responses.POST,
darwin_client.url + endpoint,
json=json_response,
status=200)
assert darwin_client.post_workflow_comment(workflow_id,
"My comment.") == comment_id
def create_dataset(dataset_slug):
""" Create new empty dataset """
client = Client.from_api_key(API_KEY)
identifier = DatasetIdentifier.parse(dataset_slug)
dataset = client.create_dataset(name=identifier.dataset_slug)
dataset_ifo = dict(
name=dataset.name,
id=dataset.dataset_id,
slug=dataset.slug,
remote_path=dataset.remote_path
)
return dataset_ifo
def authenticate(api_key: str,
default_team: Optional[bool] = None,
datasets_dir: Optional[Path] = None) -> Config:
"""
Authenticate the API key against the server and creates a configuration file for it.
Parameters
----------
api_key : str
API key to use for the client login.
default_team: Optional[bool]
Flag to make the team the default one. Defaults to None.
datasets_dir: Optional[Path]
Dataset directory on the file system. Defaults to None.
Returns
-------
Config
A configuration object to handle YAML files.
"""
# Resolve the home folder if the dataset_dir starts with ~ or ~user
validate_api_key(api_key)
try:
client = Client.from_api_key(api_key=api_key)
config_path = Path.home() / ".darwin" / "config.yaml"
config_path.parent.mkdir(exist_ok=True)
if default_team is None:
default_team = input(
f"Make {client.default_team} the default team? [y/N] ") in [
"Y", "y"
]
if datasets_dir is None:
datasets_dir = Path(
prompt("Datasets directory", "~/.darwin/datasets"))
datasets_dir = Path(datasets_dir).expanduser()
Path(datasets_dir).mkdir(parents=True, exist_ok=True)
client.set_datasets_dir(datasets_dir)
default_team_name: Optional[
str] = client.default_team if default_team else None
return persist_client_configuration(client,
default_team=default_team_name)
except InvalidLogin:
_error("Invalid API key")
def it_returns_list_of_datasets(darwin_client: Client):
team_slug: str = "v7"
endpoint: str = "/datasets"
json_response: List[Dict[str, Any]] = [
{
"name": "dataset-name-1",
"slug": "dataset-slug-1",
"id": 1,
"num_images": 1,
"num_videos": 0,
"progress": 0,
},
{
"name": "dataset-name-2",
"slug": "dataset-slug-2",
"id": 2,
"num_images": 2,
"num_videos": 0,
"progress": 0,
},
]
responses.add(responses.GET,
darwin_client.url + endpoint,
json=json_response,
status=200)
remote_datasets = list(darwin_client.list_remote_datasets(team_slug))
expected_dataset_1 = RemoteDataset(
team=team_slug,
name="dataset-name-1",
slug="dataset-slug-1",
dataset_id=1,
item_count=1,
client=darwin_client,
)
expected_dataset_2 = RemoteDataset(
team=team_slug,
name="dataset-name-2",
slug="dataset-slug-2",
dataset_id=2,
item_count=2,
client=darwin_client,
)
assert_dataset(remote_datasets[0], expected_dataset_1)
assert_dataset(remote_datasets[1], expected_dataset_2)
def _populate_dataset(dataset_slug, items):
client = Client.from_api_key(API_KEY)
identifier = DatasetIdentifier.parse(dataset_slug)
dataset = client.get_remote_dataset(dataset_identifier=identifier)
item_batches = [x.tolist() for x in np.array_split(items, min(len(items), 100))]
for idx, batch in enumerate(item_batches):
print(f'Batch {idx + 1}/{len(item_batches)}')
payload = {
'files': batch
}
print(payload)
print(dataset.dataset_id)
response = requests.put(f'https://darwin.v7labs.com/api/datasets/{dataset.dataset_id}/external_data', headers=HEADERS,
json=payload)
response.raise_for_status()
def it_returns_remote_classes(darwin_client: Client):
team_slug: str = "v7"
endpoint: str = f"/teams/{team_slug}/annotation_classes?include_tags=true"
response: Dict[str, Any] = {
"annotation_classes": [{
"annotation_class_image_url": None,
"annotation_types": ["tag"],
"dataset_id": 215,
"datasets": [{
"id": 215
}, {
"id": 265
}],
"description": " Tag 2",
"id": 345,
"images": [],
"inserted_at": "2021-01-25T02:27:10",
"metadata": {
"_color": "rgba(0,255,0,1.0)",
"tag": {}
},
"name": " Tag 2",
"team_id": 2,
"updated_at": "2021-01-25T02:27:10",
}]
}
responses.add(responses.GET,
darwin_client.url + endpoint,
json=response,
status=200)
result: List[Dict[str,
Any]] = darwin_client.fetch_remote_classes(team_slug)
annotation_class: Dict[str, Any] = result[0]
assert annotation_class["annotation_class_image_url"] is None
assert annotation_class["annotation_types"] == ["tag"]
assert annotation_class["dataset_id"] == 215
assert annotation_class["datasets"] == [{"id": 215}, {"id": 265}]
assert annotation_class["id"] == 345
def _load_client(team: Optional[str] = None, offline: bool = False):
"""Fetches a client, potentially offline
Parameters
----------
offline : bool
Flag for using an offline client
Returns
-------
Client
The client requested
"""
try:
config_dir = Path.home() / ".darwin" / "config.yaml"
client = Client.from_config(config_dir, team_slug=team)
return client
except MissingConfig:
_error("Authenticate first")
except InvalidLogin:
_error("Please re-authenticate")
except Unauthenticated:
_error("Please re-authenticate")
def it_returns_list_of_features(darwin_client: Client):
team_slug: str = "v7"
endpoint: str = f"/teams/{team_slug}/features"
json_response = [
{
"enabled": False,
"name": "WORKFLOW_V2"
},
{
"enabled": True,
"name": "BLIND_STAGE"
},
]
responses.add(responses.GET,
darwin_client.url + endpoint,
json=json_response,
status=200)
assert darwin_client.get_team_features(team_slug) == [
Feature(name="WORKFLOW_V2", enabled=False),
Feature(name="BLIND_STAGE", enabled=True),
]
def get_darwin_dataset(
*,
team_slug: Optional[str] = None,
dataset_slug: Optional[str] = None,
dataset_id: Optional[str] = None,
projects_dir: Optional[str] = None,
token: Optional[str] = None,
config_path: Optional[Path] = None,
email: Optional[str] = None,
password: Optional[str] = None,
val_percentage: Optional[float] = 0.1,
test_percentage: Optional[float] = 0.2,
force_resplit: Optional[bool] = False,
split_seed: Optional[int] = 42
):
"""
Download a Darwin dataset on the file system.
It is possible to select the way to authenticate and the configuration of
the split of the dataset
Parameters
----------
team_slug : str
Slug of the team to select
dataset_slug : str
This is the dataset name with everything lower-case, removed specials characters and
spaces are replaced by dashes, e.g., `bird-species`. This string is unique within a team
projects_dir : Path
Path where the client should be initialized from (aka the root path)
token : str
Access token used to auth a specific request. It has a time spans of roughly 8min. to
config_path : str
Path to a configuration file to use to create the client
email : str
Email of the Darwin user to use for the login
password : str
Password of the Darwin user to use for the login
val_percentage : float
Percentage of images used in the validation set
test_percentage : float
Percentage of images used in the test set
force_resplit : bool
Discard previous split and create a new one
split_seed : in
Fix seed for random split creation
Returns
-------
splits : dict
Keys are the different splits (random, tags, ...) and values are the relative file names
"""
# Authenticate client. The priority of the cases is arbitrarily chosen and should actually not matter
if email is not None and password is not None:
client = Client.login(email=email, password=password, projects_dir=projects_dir)
elif token is not None:
client = Client.from_token(token=token, projects_dir=projects_dir)
elif config_path is not None:
client = Client.from_config(config_path=config_path)
else:
client = Client.default(projects_dir=projects_dir)
# Select the desired team
if team_slug is not None:
client.set_team(slug=team_slug)
# Get the remote dataset
dataset = client.get_remote_dataset(slug=dataset_slug, dataset_id=dataset_id)
# Download the data on the file system
dataset.pull()
# Split the dataset with the param required
return split_dataset(
dataset=dataset,
val_percentage=val_percentage,
test_percentage=test_percentage,
force_resplit=force_resplit,
split_seed=split_seed
)
def get_annotations(dataset_slug, anno_dest_dir='annos', *, clear_directory=False, verbose=False):
""" Get all annotations for a dataset
dataset_name: name of the dataset to retrieve annotations for
anno_dest_dir: directory to store the annotation files
clear_directory: delete all existing files in target directory if they exist (if False raise an error if files exist)
verbose: log API responses
"""
client = Client.from_api_key(API_KEY)
identifier = DatasetIdentifier.parse(dataset_slug)
dataset = client.get_remote_dataset(dataset_identifier=identifier)
filters = {'statuses': 'review,complete'}
ids = [file.id for file in dataset.fetch_remote_files(filters)]
# darwin-py doesn't support dataset_item_ids
# uses also /datasets/{self.dataset_id}/exports
# dataset.export(annotation_class_ids=annotation_class_ids, name=name, include_url_token=include_url_token)
export_name = 'export_tmp'
print(ids)
payload = dict(
format='json',
name=export_name,
include_authorship=True,
include_export_token=True,
dataset_item_ids=ids
)
print('Creating export...')
response_create = requests.post(f'https://darwin.v7labs.com/api/teams/{TEAM_SLUG}/datasets/{dataset_slug}/exports',
headers=HEADERS,
json=payload)
response_create.raise_for_status()
if verbose:
pprint.pprint(['create_export', response_create.json()])
def get_export(timeout=60):
waiting_for_export = True
timeout_stop = time.time() + timeout
while waiting_for_export:
response_retrieve = requests.get(f'https://darwin.v7labs.com/api/teams/{TEAM_SLUG}/datasets/{dataset_slug}/exports', headers=HEADERS)
if verbose:
pprint.pprint(['get_export', response_retrieve.json()])
response_retrieve.raise_for_status()
exports = list(filter(lambda x: x['name'] == export_name, response_retrieve.json()))
if len(exports) == 1 and exports[0]['latest']:
return exports[0]
else:
if time.time() > timeout_stop:
raise RuntimeError('Timeout whilst waiting for export to complete')
time.sleep(0.5)
if verbose:
print('trying again...')
try:
print('Waiting for export to complete...')
export = get_export()
# download export data
# (this is also available through dataset.annotations as a single dict? maybe deprecated?)
print('Downloading annotations...')
with requests.get(export['download_url'], stream=True) as r:
r.raise_for_status()
if verbose:
pprint.pprint(['download_annos', r.status_code])
with mktmpdir() as tmp_dir:
tmp_file = os.path.join(tmp_dir, 'export.zip')
with open(tmp_file, 'wb') as f:
for chunk in r.iter_content(chunk_size=8192):
f.write(chunk)
if os.path.exists(anno_dest_dir):
anno_files = os.listdir(anno_dest_dir)
if len(anno_files) > 0:
if clear_directory:
for file in anno_files:
os.remove(os.path.join(anno_dest_dir, file))
else:
raise RuntimeError('Directory already exists and contains files!')
else:
os.makedirs(anno_dest_dir)
with ZipFile(tmp_file, 'r') as f:
f.extractall(anno_dest_dir)
anno_paths = [os.path.join(anno_dest_dir, x) for x in os.listdir(anno_dest_dir)]
except Exception as e:
response_delete = requests.delete(f'https://darwin.v7labs.com/api/teams/{TEAM_SLUG}/datasets/{dataset_slug}/exports/{export_name}', headers=HEADERS)
response_delete.raise_for_status()
if verbose:
pprint.pprint(['delete_export', response_delete.status_code])
raise e
print('Export completed, cleaning up...')
response_delete = requests.delete(f'https://darwin.v7labs.com/api/teams/{TEAM_SLUG}/datasets/{dataset_slug}/exports/{export_name}', headers=HEADERS)
response_delete.raise_for_status()
del export['download_url']
export['annotation_paths'] = anno_paths
return export
