def test_update_args(args, classifier_modified):
"""
Args:
args(@pytest.fixture): Namespace
classifier_modified(@pytest.fixture): ClassifierModified
Asserts: True if exp.ARGS is updated adequately.
"""
expected_classifier_args_before_update = {'dropout': 0.2, 'dim_h': 100}
expected_classifier_args_after_update = {'dropout': 0.1, 'dim_h': 100}
args_for_update = {
'data': {
'batch_size': 128
},
'model': {
'classifier_args': {
'dropout': 0.1
}
}
}
config.set_config()
classifier_modified = setup_experiment(args,
model=classifier_modified,
testmode=True)
assert exp.ARGS['model'][
'classifier_args'] == expected_classifier_args_before_update
update_args(args_for_update, exp.ARGS)
assert exp.ARGS['model'][
'classifier_args'] == expected_classifier_args_after_update
assert exp.ARGS['data']['batch_size'] == 128
def run(model=None):
'''Main function.
'''
# Parse the command-line arguments
try:
args = setup_cortex(model=model)
if args.command == 'setup':
# Performs setup only.
config.setup()
exit(0)
else:
config.set_config()
print_section('EXPERIMENT')
model, reload_nets = setup_experiment(args, model=model)
print_section('DATA')
data.setup(**exp.ARGS['data'])
print_section('MODEL')
model.reload_nets(reload_nets)
model.build()
print_section('OPTIMIZER')
optimizer.setup(model, **exp.ARGS['optimizer'])
except KeyboardInterrupt:
print('Cancelled')
exit(0)
print_section('RUNNING')
train.main_loop(model, **exp.ARGS['train'])
def test_update_nested_dicts(args, classifier_modified):
"""
Args:
args(@pytest.fixture): Namespace
classifier_modified(@pytest.fixture): ClassifierModified
Asserts: True if a dict. arg. is being updated to a
nested dict. (not overridden).
"""
expected_classifier_args_before_update = {'dropout': 0.2}
expected_classifier_args_after_update = {'dropout': 0.2, 'dim_h': 100}
args_for_update = {
'data': {
'batch_size': 122
},
'model': {
'classifier_args': {
'dropout': 0.2,
'dim_h': 100
}
}
}
config.set_config()
classifier_modified = setup_experiment(args,
model=classifier_modified,
testmode=True)
assert exp.ARGS['model'][
'classifier_args'] == expected_classifier_args_before_update
update_nested_dicts(args_for_update, exp.ARGS)
assert exp.ARGS['model'][
'classifier_args'] == expected_classifier_args_after_update
assert exp.ARGS['data']['batch_size'] == 122
def test_update_args(args, classifier_modified):
"""
Args:
args(@pytest.fixture): Namespace
classifier_modified(@pytest.fixture): ClassifierModified
Asserts: True if exp.ARGS is updated adequately.
"""
expected_classifier_args_before_update = {'dropout': 0.2}
expected_classifier_args_after_update = {'dropout': 0.1}
args_for_update = {
'data': {
'batch_size': 128
},
'model': {
'classifier_args': {
'dropout': 0.1
}
}
}
config.set_config()
classifier_modified = setup_experiment(
args, model=classifier_modified, testmode=True)
assert exp.ARGS['model'][
'classifier_args'] == expected_classifier_args_before_update
update_args(args_for_update, exp.ARGS)
assert exp.ARGS['model'][
'classifier_args'] == expected_classifier_args_after_update
assert exp.ARGS['data']['batch_size'] == 128
def test_update_nested_dicts(args, classifier_modified):
"""
Args:
args(@pytest.fixture): Namespace
classifier_modified(@pytest.fixture): ClassifierModified
Asserts: True if a dict. arg. is being updated to a
nested dict. (not overridden).
"""
expected_classifier_args_before_update = {'dropout': 0.2}
expected_classifier_args_after_update = {'dropout': 0.2, 'dim_h': 100}
args_for_update = {
'data': {
'batch_size': 122
},
'model': {
'classifier_args': {
'dropout': 0.2,
'dim_h': 100
}
}
}
config.set_config()
classifier_modified = setup_experiment(
args, model=classifier_modified, testmode=True)
assert exp.ARGS['model'][
'classifier_args'] == expected_classifier_args_before_update
update_nested_dicts(args_for_update, exp.ARGS)
assert exp.ARGS['model'][
'classifier_args'] == expected_classifier_args_after_update
assert exp.ARGS['data']['batch_size'] == 122
def run(model=None):
'''Main function.
'''
# Parse the command-line arguments
try:
args = setup_cortex(model=model)
if args.command == 'setup':
# Performs setup only.
config.setup()
exit(0)
else:
config.set_config()
print_section('EXPERIMENT')
model, reload_nets = setup_experiment(args, model=model)
print_section('DATA')
data.setup(**exp.ARGS['data'])
print_section('MODEL')
model.reload_nets(reload_nets)
model.build()
print_section('OPTIMIZER')
optimizer.setup(model, **exp.ARGS['optimizer'])
except KeyboardInterrupt:
print('Cancelled')
exit(0)
print_section('RUNNING')
train.main_loop(model, **exp.ARGS['train'])
def test_static_override_parameters(args, classifier_modified):
"""
Args:
args(@pytest.fixture): Namespace
classifier_modified(@pytest.fixture): ClassifierModified
Asserts: True if default attribute is overriding
parameters values.
"""
expected_type = 'convnet'
config.set_config()
classifier_modified = setup_experiment(
args, model=classifier_modified, testmode=True)
assert exp.ARGS['model']['classifier_type'] == expected_type
def test_static_override_parameters(args, classifier_modified):
"""
Args:
args(@pytest.fixture): Namespace
classifier_modified(@pytest.fixture): ClassifierModified
Asserts: True if default attribute is overriding
parameters values.
"""
expected_type = 'convnet'
config.set_config()
classifier_modified = setup_experiment(args,
model=classifier_modified,
testmode=True)
assert exp.ARGS['model']['classifier_type'] == expected_type
def test_command_override_static(args):
"""
Args:
args(@pytest.fixture): Namespace
Asserts: True if passing a command line arg, the exp.ARGS is
changing the value from default for the command line
one.
"""
expected_type = 'resnet'
args.__dict__['classifier_type'] = expected_type
classifier_defaults = ImageClassification()
config.set_config()
# NOTE: exp.ARGS is being populated inside setup_experiment() call
classifier_defaults = setup_experiment(
args, model=classifier_defaults, testmode=True)
assert exp.ARGS['model']['classifier_type'] == expected_type
def test_command_override_static(args):
"""
Args:
args(@pytest.fixture): Namespace
Asserts: True if passing a command line arg, the exp.ARGS is
changing the value from default for the command line
one.
"""
expected_type = 'resnet'
args.__dict__['classifier_type'] = expected_type
classifier_defaults = ImageClassification()
config.set_config()
# NOTE: exp.ARGS is being populated inside setup_experiment() call
classifier_defaults = setup_experiment(args,
model=classifier_defaults,
testmode=True)
assert exp.ARGS['model']['classifier_type'] == expected_type