def test_import_all_models():
""" Try loading all defined models to ensure that their full qualified
names are still good
"""
for model_name in MODELS.keys():
print("Try loading model", model_name)
get_model_class(model_name)
def test_import_all_models():
"""Try loading all defined models to ensure that their full qualified
names are still good
"""
for model_name in MODELS:
if model_name == "component_nn" and not importlib.util.find_spec(
"tensorflow"):
continue
print("Try loading model", model_name)
get_model_class(model_name)
def go(self, model_name):
# Download datasets that were built by bugbug_data.
os.makedirs("data", exist_ok=True)
model_class = get_model_class(model_name)
if issubclass(model_class, model.BugModel) or issubclass(
model_class, model.BugCoupleModel
):
self.download_db("bugs")
if issubclass(model_class, model.CommitModel):
self.download_db("commits")
logger.info(f"Training *{model_name}* model")
model_obj = model_class()
model_obj.train()
logger.info(f"Training done")
model_file_name = f"{model_name}model"
assert os.path.exists(model_file_name)
self.compress_file(model_file_name)
logger.info(f"Model compressed")
def main(args):
if args.goal == "component":
if args.classifier == "default":
model_class_name = "component"
else:
model_class_name = "component_nn"
else:
model_class_name = args.goal
model_class = get_model_class(model_class_name)
if args.train:
db.download(bugzilla.BUGS_DB)
db.download(repository.COMMITS_DB)
historical_supported_tasks = [
"defect",
"bugtype",
"defectenhancementtask",
"regression",
]
if args.goal in historical_supported_tasks:
model = model_class(args.lemmatization, args.historical)
elif args.goal == "duplicate":
model = model_class(
args.training_set_size, args.lemmatization, args.cleanup_urls
)
else:
model = model_class(args.lemmatization)
model.train()
def go(self, model_name):
# Download datasets that were built by bugbug_data.
os.makedirs("data", exist_ok=True)
model_class = get_model_class(model_name)
model_obj = model_class()
if (isinstance(model_obj, model.BugModel)
or isinstance(model_obj, model.BugCoupleModel)
or (hasattr(model_obj, "bug_data") and model_obj.bug_data)):
db.download(bugzilla.BUGS_DB, force=True)
if isinstance(model_obj, model.CommitModel):
db.download(repository.COMMITS_DB, force=True)
logger.info(f"Training *{model_name}* model")
metrics = model_obj.train()
# Save the metrics as a file that can be uploaded as an artifact.
metric_file_path = "metrics.json"
with open(metric_file_path, "w") as metric_file:
json.dump(metrics, metric_file, cls=CustomJsonEncoder)
logger.info(f"Training done")
model_file_name = f"{model_name}model"
assert os.path.exists(model_file_name)
zstd_compress(model_file_name)
logger.info(f"Model compressed")
def classify_bugs(model_name, classifier, bug_id):
if classifier != "default":
assert (
model_name in MODELS_WITH_TYPE
), f"{classifier} is not a valid classifier type for {model_name}"
model_file_name = f"{model_name}{classifier}model"
model_name = f"{model_name}_{classifier}"
else:
model_file_name = f"{model_name}model"
if not os.path.exists(model_file_name):
logger.info(f"{model_file_name} does not exist. Downloading the model....")
try:
download_check_etag(
f"https://index.taskcluster.net/v1/task/project.relman.bugbug.train_{model_name}.latest/artifacts/public/{model_file_name}.zst",
f"{model_file_name}.zst",
)
except requests.HTTPError:
logger.error(
f"A pre-trained model is not available, you will need to train it yourself using the trainer script"
)
raise SystemExit(1)
zstd_decompress(model_file_name)
assert os.path.exists(model_file_name), "Decompressed file doesn't exist"
model_class = get_model_class(model_name)
model = model_class.load(model_file_name)
if bug_id:
bugs = bugzilla.get(bug_id).values()
assert bugs, f"A bug with a bug id of {bug_id} was not found"
else:
bugs = bugzilla.get_bugs()
for bug in bugs:
print(
f'https://bugzilla.mozilla.org/show_bug.cgi?id={bug["id"]} - {bug["summary"]} '
)
if model.calculate_importance:
probas, importance = model.classify(
bug, probabilities=True, importances=True
)
model.print_feature_importances(
importance["importances"], class_probabilities=probas
)
with open("importance.html", "w") as f:
f.write(importance["html"])
else:
probas = model.classify(bug, probabilities=True, importances=False)
if np.argmax(probas) == 1:
print(f"Positive! {probas}")
else:
print(f"Negative! {probas}")
input()
def go(self, model_name):
# Download datasets that were built by bugbug_data.
os.makedirs("data", exist_ok=True)
model_class = get_model_class(model_name)
if issubclass(model_class, model.BugModel) or issubclass(
model_class, model.BugCoupleModel
):
self.download_db("bugs")
if issubclass(model_class, model.CommitModel):
self.download_db("commits")
logger.info(f"Training *{model_name}* model")
model_obj = model_class()
metrics = model_obj.train()
# Save the metrics as a file that can be uploaded as an artifact.
metric_file_path = "metrics.json"
with open(metric_file_path, "w") as metric_file:
json.dump(metrics, metric_file, cls=CustomJsonEncoder)
logger.info(f"Training done")
model_file_name = f"{model_name}model"
assert os.path.exists(model_file_name)
self.compress_file(model_file_name)
logger.info(f"Model compressed")
def go(self, args):
# Download datasets that were built by bugbug_data.
os.makedirs("data", exist_ok=True)
if args.classifier != "default":
assert (
args.model in MODELS_WITH_TYPE
), f"{args.classifier} is not a valid classifier type for {args.model}"
model_name = f"{args.model}_{args.classifier}"
else:
model_name = args.model
model_class = get_model_class(model_name)
if args.model in HISTORICAL_SUPPORTED_TASKS:
model_obj = model_class(args.lemmatization, args.historical)
elif args.model == "duplicate":
model_obj = model_class(args.training_set_size, args.lemmatization,
args.cleanup_urls)
else:
model_obj = model_class(args.lemmatization)
if (isinstance(model_obj, model.BugModel)
or isinstance(model_obj, model.BugCoupleModel)
or (hasattr(model_obj, "bug_data") and model_obj.bug_data)):
if args.download_db:
db.download(bugzilla.BUGS_DB)
else:
logger.info("Skipping download of the bug database")
if isinstance(model_obj, model.CommitModel):
if args.download_db:
db.download(repository.COMMITS_DB)
else:
logger.info("Skipping download of the commit database")
logger.info(f"Training *{model_name}* model")
metrics = model_obj.train(limit=args.limit)
# Save the metrics as a file that can be uploaded as an artifact.
metric_file_path = "metrics.json"
with open(metric_file_path, "w") as metric_file:
json.dump(metrics, metric_file, cls=CustomJsonEncoder)
logger.info(f"Training done")
model_file_name = f"{model_name}model"
assert os.path.exists(model_file_name)
zstd_compress(model_file_name)
logger.info(f"Model compressed")
if model_obj.store_dataset:
assert os.path.exists(f"{model_file_name}_data_X")
zstd_compress(f"{model_file_name}_data_X")
assert os.path.exists(f"{model_file_name}_data_y")
zstd_compress(f"{model_file_name}_data_y")
def load_model(self, model_name):
model_path = f"{model_name}model"
if not os.path.exists(model_path):
download_check_etag(
URL.format(model_name=model_name,
file_name=f"{model_path}.zst"))
zstd_decompress(model_path)
assert os.path.exists(model_path), "Decompressed model exists"
return get_model_class(model_name).load(model_path)
def go(self, args):
# Download datasets that were built by bugbug_data.
os.makedirs("data", exist_ok=True)
if args.classifier != "default":
assert (
args.model in MODELS_WITH_TYPE
), f"{args.classifier} is not a valid classifier type for {args.model}"
model_name = f"{args.model}_{args.classifier}"
else:
model_name = args.model
model_class = get_model_class(model_name)
if args.model in HISTORICAL_SUPPORTED_TASKS:
model_obj = model_class(args.lemmatization, args.historical)
elif args.model == "regressor":
model_obj = model_class(args.lemmatization, args.interpretable)
elif args.model == "duplicate":
model_obj = model_class(
args.training_set_size, args.lemmatization, args.cleanup_urls
)
else:
model_obj = model_class(args.lemmatization)
if args.download_db:
for required_db in model_obj.training_dbs:
assert db.download(required_db)
if args.download_eval:
model_obj.download_eval_dbs()
else:
logger.info("Skipping download of the databases")
logger.info(f"Training *{model_name}* model")
metrics = model_obj.train(limit=args.limit)
# Save the metrics as a file that can be uploaded as an artifact.
metric_file_path = "metrics.json"
with open(metric_file_path, "w") as metric_file:
json.dump(metrics, metric_file, cls=CustomJsonEncoder)
logger.info("Training done")
model_file_name = f"{model_name}model"
assert os.path.exists(model_file_name)
zstd_compress(model_file_name)
logger.info("Model compressed")
if model_obj.store_dataset:
assert os.path.exists(f"{model_file_name}_data_X")
zstd_compress(f"{model_file_name}_data_X")
assert os.path.exists(f"{model_file_name}_data_y")
zstd_compress(f"{model_file_name}_data_y")
def classify_bugs(model_name: str, classifier: str, bug_id: int) -> None:
if classifier != "default":
assert (
model_name in MODELS_WITH_TYPE
), f"{classifier} is not a valid classifier type for {model_name}"
model_file_name = f"{model_name}{classifier}model"
model_name = f"{model_name}_{classifier}"
else:
model_file_name = f"{model_name}model"
if not os.path.exists(model_file_name):
logger.info(
f"{model_file_name} does not exist. Downloading the model....")
try:
download_model(model_name)
except requests.HTTPError:
logger.error(
"A pre-trained model is not available, you will need to train it yourself using the trainer script"
)
raise SystemExit(1)
model_class = get_model_class(model_name)
model = model_class.load(model_file_name)
if bug_id:
bugs = bugzilla.get(bug_id).values()
assert bugs, f"A bug with a bug id of {bug_id} was not found"
else:
assert db.download(bugzilla.BUGS_DB)
bugs = bugzilla.get_bugs()
for bug in bugs:
print(
f'https://bugzilla.mozilla.org/show_bug.cgi?id={bug["id"]} - {bug["summary"]} '
)
if model.calculate_importance:
probas, importance = model.classify(bug,
probabilities=True,
importances=True)
model.print_feature_importances(importance["importances"],
class_probabilities=probas)
else:
probas = model.classify(bug, probabilities=True, importances=False)
probability = probas[0]
pred_index = np.argmax(probability)
if len(probability) > 2:
pred_class = model.le.inverse_transform([pred_index])[0]
else:
pred_class = "Positive" if pred_index == 1 else "Negative"
print(f"{pred_class} {probability}")
input()
def classify_issues(owner: str, repo: str, retrieve_events: bool,
model_name: str, issue_number: int) -> None:
model_file_name = f"{model_name}model"
if not os.path.exists(model_file_name):
logger.info(
f"{model_file_name} does not exist. Downloading the model....")
try:
download_model(model_name)
except requests.HTTPError:
logger.error(
"A pre-trained model is not available, you will need to train it yourself using the trainer script"
)
raise SystemExit(1)
model_class = get_model_class(model_name)
model = model_class.load(model_file_name)
if issue_number:
issues = iter([
github.fetch_issue_by_number(owner, repo, issue_number,
retrieve_events)
])
assert issues, f"An issue with a number of {issue_number} was not found"
else:
assert db.download(github.GITHUB_ISSUES_DB)
issues = github.get_issues()
for issue in issues:
print(f'{issue["url"]} - {issue["title"]} ')
if model.calculate_importance:
probas, importance = model.classify(issue,
probabilities=True,
importances=True)
model.print_feature_importances(importance["importances"],
class_probabilities=probas)
else:
probas = model.classify(issue,
probabilities=True,
importances=False)
probability = probas[0]
pred_index = np.argmax(probability)
if len(probability) > 2:
pred_class = model.le.inverse_transform([pred_index])[0]
else:
pred_class = "Positive" if pred_index == 1 else "Negative"
print(f"{pred_class} {probability}")
input()
def go(self, args):
# Download datasets that were built by bugbug_data.
os.makedirs("data", exist_ok=True)
if args.classifier != "default":
assert (
args.model in MODELS_WITH_TYPE
), f"{args.classifier} is not a valid classifier type for {args.model}"
model_name = f"{args.model}_{args.classifier}"
else:
model_name = args.model
model_class = get_model_class(model_name)
parameter_names = set(
inspect.signature(model_class.__init__).parameters)
parameters = {
key: value
for key, value in vars(args).items() if key in parameter_names
}
model_obj = model_class(**parameters)
if args.download_db:
for required_db in model_obj.training_dbs:
assert db.download(required_db)
if args.download_eval:
model_obj.download_eval_dbs()
else:
logger.info("Skipping download of the databases")
logger.info(f"Training *{model_name}* model")
metrics = model_obj.train(limit=args.limit)
# Save the metrics as a file that can be uploaded as an artifact.
metric_file_path = "metrics.json"
with open(metric_file_path, "w") as metric_file:
json.dump(metrics, metric_file, cls=CustomJsonEncoder)
logger.info("Training done")
model_file_name = f"{model_name}model"
assert os.path.exists(model_file_name)
zstd_compress(model_file_name)
logger.info("Model compressed")
if model_obj.store_dataset:
assert os.path.exists(f"{model_file_name}_data_X")
zstd_compress(f"{model_file_name}_data_X")
assert os.path.exists(f"{model_file_name}_data_y")
zstd_compress(f"{model_file_name}_data_y")
def generate_sheet(model_name, token, days, threshold):
model_file_name = f"{model_name}model"
assert os.path.exists(
model_file_name
), f"{model_file_name} does not exist. Train the model with trainer.py first."
model_class = get_model_class(model_name)
model = model_class.load(model_file_name)
today = datetime.utcnow()
start_date = today - timedelta(days)
bugzilla.set_token(token)
bug_ids = bugzilla.get_ids_between(start_date, today)
bugs = bugzilla.get(bug_ids)
print(f"Classifying {len(bugs)} bugs...")
rows = [["Bug", f"{model_name}(model)", model_name, "Title"]]
for bug in bugs.values():
p = model.classify(bug, probabilities=True)
probability = p[0]
if len(probability) > 2:
index = np.argmax(probability)
prediction = model.class_names[index]
else:
prediction = "y" if probability[1] >= threshold else "n"
rows.append(
[
f'https://bugzilla.mozilla.org/show_bug.cgi?id={bug["id"]}',
prediction,
"",
bug["summary"],
]
)
os.makedirs("sheets", exist_ok=True)
with open(
os.path.join(
"sheets",
f'{model_name}-{datetime.utcnow().strftime("%Y-%m-%d")}-labels.csv',
),
"w",
) as f:
writer = csv.writer(f)
writer.writerows(rows)
def classify_bugs(model_name, classifier):
if classifier != "default":
assert (
model_name in MODELS_WITH_TYPE
), f"{classifier} is not a valid classifier type for {model_name}"
model_file_name = f"{model_name}{classifier}model"
model_name = f"{model_name}_{classifier}"
else:
model_file_name = f"{model_name}model"
assert os.path.exists(
model_file_name
), f"{model_file_name} does not exist. Train the model with trainer.py first."
model_class = get_model_class(model_name)
model = model_class.load(model_file_name)
for bug in bugzilla.get_bugs():
print(
f'https://bugzilla.mozilla.org/show_bug.cgi?id={bug["id"]} - {bug["summary"]} '
)
if model.calculate_importance:
probas, importance = model.classify(
bug, probabilities=True, importances=True
)
feature_names = model.get_human_readable_feature_names()
model.print_feature_importances(
importance["importances"], feature_names, class_probabilities=probas
)
else:
probas = model.classify(bug, probabilities=True, importances=False)
if np.argmax(probas) == 1:
print(f"Positive! {probas}")
else:
print(f"Negative! {probas}")
input()
def generate_sheet(model_name, token):
model_file_name = f"{model_name}model"
assert os.path.exists(
model_file_name
), f"{model_file_name} does not exist. Train the model with trainer.py first."
model_class = get_model_class(model_name)
model = model_class.load(model_file_name)
today = datetime.utcnow()
a_week_ago = today - timedelta(7)
bugzilla.set_token(token)
bug_ids = bugzilla.get_ids_between(a_week_ago, today)
bugs = bugzilla.get(bug_ids)
print(f"Classifying {len(bugs)} bugs...")
rows = [["Bug", f"{model_name}(model)", model_name, "Title"]]
for bug in bugs.values():
p = model.classify(bug, probabilities=True)
rows.append([
f'https://bugzilla.mozilla.org/show_bug.cgi?id={bug["id"]}',
"y" if p[0][1] >= 0.7 else "n",
"",
bug["summary"],
])
os.makedirs("sheets", exist_ok=True)
with open(
os.path.join(
"sheets",
f'{model_name}-{datetime.utcnow().strftime("%Y-%m-%d")}-labels.csv',
),
"w",
) as f:
writer = csv.writer(f)
writer.writerows(rows)
def go(self, model_name):
# Download datasets that were built by bugbug_data.
os.makedirs("data", exist_ok=True)
# Bugs.json
logger.info("Downloading bugs database")
bugs_url = BASE_URL.format("bugs")
urlretrieve(f"{bugs_url}/bugs.json.xz", "data/bugs.json.xz")
logger.info("Decompressing bugs database")
self.decompress_file("data/bugs.json")
logger.info(f"Training *{model_name}* model")
model_class = get_model_class(model_name)
model = model_class()
model.train()
logger.info(f"Training done")
model_file_name = f"{model_name}model"
assert os.path.exists(model_file_name)
self.compress_file(model_file_name)
logger.info(f"Model compressed")
args = parser.parse_args()
model_file_name = "{}{}model".format(
args.goal, "" if args.classifier == "default" else args.classifier)
model_class_name = args.goal
if args.goal == "component":
if args.classifier == "default":
model_class_name = "component"
elif args.classifier == "nn":
model_class_name = "component_nn"
else:
raise ValueError(f"Unknown value {args.classifier}")
model_class = get_model_class(model_class_name)
if args.train:
db.download(bugzilla.BUGS_DB)
db.download(repository.COMMITS_DB)
if args.historical:
model = model_class(args.lemmatization, args.historical)
else:
model = model_class(args.lemmatization)
model.train()
else:
model = model_class.load(model_file_name)
if args.classify:
for bug in bugzilla.get_bugs():
def download_and_load_model(model_name):
path = download_model(model_name)
return get_model_class(model_name).load(path)
def main(args):
model_file_name = "{}{}model".format(
args.goal, "" if args.classifier == "default" else args.classifier)
if args.goal == "component":
if args.classifier == "default":
model_class_name = "component"
else:
model_class_name = "component_nn"
else:
model_class_name = args.goal
model_class = get_model_class(model_class_name)
if args.train:
db.download(bugzilla.BUGS_DB)
db.download(repository.COMMITS_DB)
historical_supported_tasks = [
"defect",
"bugtype",
"defectenhancementtask",
"regression",
]
if args.goal in historical_supported_tasks:
model = model_class(args.lemmatization, args.historical)
elif args.goal == "duplicate":
model = model_class(args.training_set_size, args.lemmatization)
else:
model = model_class(args.lemmatization)
model.train()
else:
model = model_class.load(model_file_name)
if args.classify:
for bug in bugzilla.get_bugs():
print(
f'https://bugzilla.mozilla.org/show_bug.cgi?id={ bug["id"] } - { bug["summary"]} '
)
if model.calculate_importance:
probas, importance = model.classify(bug,
probabilities=True,
importances=True)
feature_names = model.get_feature_names()
for i, (importance, index,
is_positive) in enumerate(importance["importances"]):
print(
f'{i + 1}. \'{feature_names[int(index)]}\' ({"+" if (is_positive) else "-"}{importance})'
)
else:
probas = model.classify(bug,
probabilities=True,
importances=False)
if np.argmax(probas) == 1:
print(f"Positive! {probas}")
else:
print(f"Negative! {probas}")
input()
if args.generate_sheet:
assert (args.token is not None
), "A Bugzilla token should be set in order to download bugs"
today = datetime.utcnow()
a_week_ago = today - timedelta(7)
bugzilla.set_token(args.token)
bugs = bugzilla.download_bugs_between(a_week_ago, today)
print(f"Classifying {len(bugs)} bugs...")
rows = [["Bug", f"{args.goal}(model)", args.goal, "Title"]]
for bug in bugs:
p = model.classify(bug, probabilities=True)
rows.append([
f'https://bugzilla.mozilla.org/show_bug.cgi?id={bug["id"]}',
"y" if p[0][1] >= 0.7 else "n",
"",
bug["summary"],
])
os.makedirs("sheets", exist_ok=True)
with open(
os.path.join(
"sheets",
f'{args.goal}-{datetime.utcnow().strftime("%Y-%m-%d")}-labels.csv',
),
"w",
) as f:
writer = csv.writer(f)
writer.writerows(rows)
def test_component_is_bugmodel():
model_class = get_model_class("component")
assert issubclass(model_class, model.BugModel)
model_class = get_model_class("regression")
assert issubclass(model_class, model.BugModel)
def test_backout_is_commitmodel():
model_class = get_model_class("backout")
assert issubclass(model_class, model.CommitModel)
def main(args):
model_file_name = "{}{}model".format(
args.goal, "" if args.classifier == "default" else args.classifier)
if args.goal == "component":
if args.classifier == "default":
model_class_name = "component"
else:
model_class_name = "component_nn"
else:
model_class_name = args.goal
model_class = get_model_class(model_class_name)
if args.train:
db.download(bugzilla.BUGS_DB)
db.download(repository.COMMITS_DB)
historical_supported_tasks = [
"defect",
"bugtype",
"defectenhancementtask",
"regression",
]
if args.goal in historical_supported_tasks:
model = model_class(args.lemmatization, args.historical)
elif args.goal == "duplicate":
model = model_class(args.training_set_size, args.lemmatization,
args.cleanup_urls)
else:
model = model_class(args.lemmatization)
model.train()
else:
model = model_class.load(model_file_name)
if args.classify:
for bug in bugzilla.get_bugs():
print(
f'https://bugzilla.mozilla.org/show_bug.cgi?id={ bug["id"] } - { bug["summary"]} '
)
if model.calculate_importance:
probas, importance = model.classify(bug,
probabilities=True,
importances=True)
feature_names = model.get_human_readable_feature_names()
model.print_feature_importances(importance["importances"],
feature_names,
class_probabilities=probas)
else:
probas = model.classify(bug,
probabilities=True,
importances=False)
if np.argmax(probas) == 1:
print(f"Positive! {probas}")
else:
print(f"Negative! {probas}")
input()
def __init__(self, model_name, cache_root, git_repo_dir,
method_defect_predictor_dir):
self.model_name = model_name
self.cache_root = cache_root
assert os.path.isdir(
cache_root), f"Cache root {cache_root} is not a dir."
self.repo_dir = os.path.join(cache_root, "mozilla-central")
model_path = f"{model_name}model"
if not os.path.exists(model_path):
download_check_etag(URL.format(model_name=model_name),
f"{model_path}.zst")
zstd_decompress(model_path)
assert os.path.exists(model_path), "Decompressed model exists"
self.model = get_model_class(model_name).load(model_path)
self.git_repo_dir = git_repo_dir
if git_repo_dir:
self.clone_git_repo("https://github.com/mozilla/gecko-dev",
git_repo_dir)
self.method_defect_predictor_dir = method_defect_predictor_dir
if method_defect_predictor_dir:
self.clone_git_repo(
"https://github.com/lucapascarella/MethodDefectPredictor",
method_defect_predictor_dir,
"fa5269b959d8ddf7e97d1e92523bb64c17f9bbcd",
)
if model_name == "regressor":
self.use_test_history = False
model_data_X_path = f"{model_name}model_data_X"
if not os.path.exists(model_data_X_path):
download_check_etag(
URL.format(f"{model_data_X_path}.zst"),
f"{model_data_X_path}.zst",
)
zstd_decompress(model_data_X_path)
assert os.path.exists(
model_data_X_path), "Decompressed X dataset exists"
model_data_y_path = f"{model_name}model_data_y"
if not os.path.exists(model_data_y_path):
download_check_etag(
URL.format(f"{model_data_y_path}.zst"),
f"{model_data_y_path}.zst",
)
zstd_decompress(model_data_y_path)
assert os.path.exists(
model_data_y_path), "Decompressed y dataset exists"
self.X = to_array(joblib.load(model_data_X_path))
self.y = to_array(joblib.load(model_data_y_path))
if model_name == "testselect":
self.use_test_history = True
assert db.download_support_file(test_scheduling.TEST_SCHEDULING_DB,
test_scheduling.PAST_FAILURES_DB)
self.past_failures_data = test_scheduling.get_past_failures()
def parse_args(args):
description = "Train the models"
main_parser = argparse.ArgumentParser(description=description)
parser = argparse.ArgumentParser(add_help=False)
parser.add_argument(
"--limit",
type=int,
help="Only train on a subset of the data, used mainly for integrations tests",
)
parser.add_argument(
"--no-download",
action="store_false",
dest="download_db",
help="Do not download databases, uses whatever is on disk",
)
parser.add_argument(
"--download-eval",
action="store_true",
dest="download_eval",
help="Download databases and database support files required at runtime (e.g. if the model performs custom evaluations)",
)
parser.add_argument(
"--lemmatization",
help="Perform lemmatization (using spaCy)",
action="store_true",
)
parser.add_argument(
"--classifier",
help="Type of the classifier. Only used for component classification.",
choices=["default", "nn"],
default="default",
)
subparsers = main_parser.add_subparsers(title="model", dest="model", required=True)
for model_name in MODELS:
subparser = subparsers.add_parser(
model_name, parents=[parser], help=f"Train {model_name} model"
)
try:
model_class_init = get_model_class(model_name).__init__
except ImportError:
continue
for parameter in inspect.signature(model_class_init).parameters.values():
if parameter.name == "self":
continue
# Skip parameters handled by the base class (TODO: add them to the common argparser and skip them automatically without hardcoding by inspecting the base class)
if parameter.name == "lemmatization":
continue
parameter_type = parameter.annotation
if parameter_type == inspect._empty:
parameter_type = type(parameter.default)
assert parameter_type is not None
if parameter_type == bool:
subparser.add_argument(
f"--{parameter.name}"
if parameter.default is False
else f"--no-{parameter.name}",
action="store_true"
if parameter.default is False
else "store_false",
dest=parameter.name,
)
else:
subparser.add_argument(
f"--{parameter.name}",
default=parameter.default,
dest=parameter.name,
type=int,
)
return main_parser.parse_args(args)
