def test_munchify_works():
def main(cfg, _):
print(cfg.attr)
dir_path = os.path.dirname(os.path.realpath(__file__))
config_path = os.path.join(dir_path, 'config2.yml')
# https://stackoverflow.com/a/37343818
with mock.patch('argparse.ArgumentParser.parse_args',
return_value=argparse.Namespace(config=config_path,
level='info')):
run(main, __file__)
def test_run_config(config_name, level, called, main):
dir_path = os.path.dirname(os.path.realpath(__file__))
config_path = os.path.join(dir_path, config_name)
# https://stackoverflow.com/a/37343818
with mock.patch('argparse.ArgumentParser.parse_args',
return_value=argparse.Namespace(config=config_path,
level='info')):
# https://stackoverflow.com/a/31756485
with mock.patch(f'logging.Logger.{level}') as mock_logger:
run(main, __file__)
mock_logger.assert_called_with(called)
def test_if_name_not_main_then_not_called():
'''
Call `run` on a method which throws an exception, after patching
`parse_args` to return None. This verifies no argument parsing is done, and
that `main` is not called.
'''
def main(cfg, log):
raise Exception('I should not be called')
# https://stackoverflow.com/a/534847
file = str(uuid.uuid4())
with mock.patch('argparse.ArgumentParser.parse_args', return_value=None):
run(main, file, '__not_main__')
def test_test_set_logger_levels_from_config_file(file, levels):
def main(cfg, _):
pass
dir_path = os.path.dirname(os.path.realpath(__file__))
config_path = os.path.join(dir_path, file)
# https://stackoverflow.com/a/37343818
with mock.patch('argparse.ArgumentParser.parse_args',
return_value=argparse.Namespace(config=config_path,
level='info')):
run(main, __file__)
for k, v in levels.items():
assert v == logging.getLogger(k).level
from azureml.core.run import Run
import model
from pyconfigurableml.entry import run
import pytorch_lightning as pl
from pytorch_lightning.utilities import rank_zero_only
from pytorch_lightning.loggers import LightningLoggerBase
from pytorch_lightning.loggers.tensorboard import TensorBoardLogger
# TODO: long-term, get this from pytorch-lightning.
from _azureml import AzureMlLogger
def main(config, log):
loggers = [
AzureMlLogger(),
TensorBoardLogger('lightning_logs')
]
trainer = pl.Trainer(logger=loggers, **config.trainer)
net = model.Mnist(**config.model)
trainer.fit(net)
run(main, __file__, __name__)
def test_run_type_checking(input) -> None:
with pytest.raises(TypeError):
run(input, __file__)