def train_command(data_path: str, forecaster: Optional[str]) -> None:
from gluonts.shell import train
logger.info("Run 'train' command")
try:
env = TrainEnv(Path(data_path))
if forecaster is None:
try:
forecaster = env.hyperparameters["forecaster_name"]
except KeyError:
msg = (
"Forecaster shell parameter is `None`, but "
"the `forecaster_name` key is not defined in the "
"hyperparameters.json dictionary."
)
raise GluonTSForecasterNotFoundError(msg)
train.run_train_and_test(env, forecaster_type_by_name(forecaster))
except Exception as error:
with open(
TrainPaths(Path(data_path)).output / "failure", "w"
) as out_file:
out_file.write(str(error))
out_file.write("\n\n")
out_file.write(traceback.format_exc())
raise
def forecaster_type_by_name(name: str) -> Forecaster:
"""
Loads a forecaster from the `gluonts_forecasters` entry_points namespace
by name.
If a forecater wasn't register under that name, it tries to locate the
class.
Third-party libraries can register their forecasters as follows by defining
a corresponding section in the `entry_points` section of their `setup.py`::
entry_points={
'gluonts_forecasters': [
'model_a = my_models.model_a:MyEstimator',
'model_b = my_models.model_b:MyPredictor',
]
}
"""
forecaster = None
for entry_point in pkg_resources.iter_entry_points("gluonts_forecasters"):
if entry_point.name == name:
forecaster = entry_point.load()
break
else:
forecaster = pydoc.locate(name)
if forecaster is None:
raise GluonTSForecasterNotFoundError(
f'Cannot locate estimator with classname "{name}".')
return cast(Forecaster, forecaster)
def forecaster_type_by_name(name: str) -> Type[Union[Estimator, Predictor]]:
"""
Loads a forecaster from the `gluonts_forecasters` entry_points namespace
by name.
Third-party libraries can register their forecasters as follows by defining
a corresponding section in the `entry_points` section of their `setup.py`::
entry_points={
'blogtool.parsers': [
'model_a = my_models.model_a:MyEstimator',
'model_b = my_models.model_b:MyPredictor',
]
}
"""
forecaster = None
for entry_point in pkg_resources.iter_entry_points('gluonts_forecasters'):
if entry_point.name == name:
forecaster = entry_point.load()
break
if forecaster is None:
msg = f'Cannot locate estimator with classname "{name}".'
raise GluonTSForecasterNotFoundError(msg)
return cast(Type[Union[Estimator, Predictor]], forecaster)
def train_command(data_path: str, forecaster: Optional[str]) -> None:
from gluonts.shell import train
env = TrainEnv(Path(data_path))
if forecaster is None:
try:
forecaster = env.hyperparameters['forecaster_name']
except KeyError:
msg = ("Forecaster shell parameter is `None`, but "
"the `forecaster_name` key is not defined in the "
"hyperparameters.json dictionary.")
raise GluonTSForecasterNotFoundError(msg)
assert forecaster is not None
train.run_train_and_test(env, forecaster_type_by_name(forecaster))
def train_command(data_path: str, forecaster: str) -> None:
from gluonts.shell import train
env = SageMakerEnv(Path(data_path))
if forecaster == '%from_hyperparameters%':
try:
forecaster = env.hyperparameters['forecaster_name']
except KeyError:
msg = (
"Forecaster shell parameter is '%from_hyperparameters%', but "
"the `forecaster_name` key is not defined in the "
"hyperparameters.json dictionary."
)
raise GluonTSForecasterNotFoundError(msg)
forecaster_type = forecaster_type_by_name(forecaster)
train.run_train_and_test(env, forecaster_type)
