def test_legacy_serialization(valid_simple_ml_predictor_data):
"""Ensure that a serialized SimplePredictor looks sane."""
predictor = SimpleMLPredictor.build(valid_simple_ml_predictor_data)
serialized = predictor.dump()
serialized['id'] = valid_simple_ml_predictor_data['id']
assert serialized == valid_serialization_output(
valid_simple_ml_predictor_data)
def test_register(valid_simple_ml_predictor_data, basic_predictor_report_data):
session = mock.Mock()
session.post_resource.return_value = valid_simple_ml_predictor_data
session.get_resource.return_value = basic_predictor_report_data
pc = PredictorCollection(uuid.uuid4(), session)
predictor = SimpleMLPredictor.build(valid_simple_ml_predictor_data)
registered = pc.register(predictor)
assert registered.name == 'ML predictor'
def test_failed_register(valid_simple_ml_data):
session = mock.Mock()
session.post_resource.side_effect = NotFound("/projects/uuid/not_found")
pc = PredictorCollection(uuid.uuid4(), session)
predictor = SimpleMLPredictor.build(valid_simple_ml_data)
with pytest.raises(ModuleRegistrationFailedException) as e:
pc.register(predictor)
assert 'The "SimpleMLPredictor" failed to register. NotFound: /projects/uuid/not_found' in str(e.value)
def simple_predictor() -> SimpleMLPredictor:
"""Build a SimpleMLPredictor for testing."""
return SimpleMLPredictor(
name='ML predictor',
description='Predicts z from input x and latent variable y',
inputs=[x],
outputs=[z],
latent_variables=[y],
training_data=[data_source])
def test_create_or_update_raise_error_multiple_found(predictor_collection):
# test when there are multiple resources that exists with the same listed name and raise error
pred = SimpleMLPredictor(
name="resource 1", #Not unique: two "resource 1" exists in collection
description='I am updated!',
inputs=[],
outputs=[],
latent_variables=[])
with pytest.raises(ValueError):
create_or_update(predictor_collection, pred)
def predictor_collection() -> PredictorCollection:
class SeedingTestPredictorCollection(PredictorCollection):
predictors = []
def register(self, model: SimpleMLPredictor) -> SimpleMLPredictor:
self.predictors.append(model)
return model
def update(self, model):
self.predictors = [
r for r in self.predictors if r.uid != model.uid
]
return self.register(model)
def list(self, page: Optional[int] = None, per_page: int = 100):
if page is None:
return self.predictors
else:
return self.predictors[(page - 1) * per_page:page * per_page]
predictors = SeedingTestPredictorCollection(
UUID('6b608f78-e341-422c-8076-35adc8828545'), session)
#Adding a few predictors in the collection to have something to update
for i in range(0, 5):
predictors.register(
SimpleMLPredictor(name="resource " + str(i),
description='',
inputs=[],
outputs=[],
latent_variables=[]))
#Adding a few predictors with the same name ("resource {0,1}" were made above)
# this is used to test behavior if there are duplicates
for i in range(0, 2):
predictors.register(
SimpleMLPredictor(name="resource " + str(i),
description='',
inputs=[],
outputs=[],
latent_variables=[]))
return predictors
def test_register(valid_simple_ml_data):
session = mock.Mock()
session.post_resource.return_value = valid_simple_ml_data
session.get_resource.return_value = {
'id': str(uuid.uuid4()),
'status': 'VALID',
'report': {}
}
pc = PredictorCollection(uuid.uuid4(), session)
predictor = SimpleMLPredictor.build(valid_simple_ml_data)
registered = pc.register(predictor)
assert registered.name == 'ML predictor'
def test_simple_legacy_deserialization(valid_simple_ml_data):
"""Ensure that a deserialized SimplePredictor looks sane."""
predictor: SimpleMLPredictor = SimpleMLPredictor.build(
valid_simple_ml_data)
assert predictor.name == 'ML predictor'
assert predictor.description == 'Predicts z from input x and latent variable y'
assert len(predictor.inputs) == 1
assert predictor.inputs[0] == x
assert len(predictor.outputs) == 1
assert predictor.outputs[0] == z
assert len(predictor.latent_variables) == 1
assert predictor.latent_variables[0] == y
assert predictor.training_data == 'training_data_key'
def test_register_experimental(valid_simple_ml_predictor_data, basic_predictor_report_data):
session = mock.Mock()
post_response = deepcopy(valid_simple_ml_predictor_data)
post_response["experimental"] = True
post_response["experimental_reasons"] = ["This is a test", "Of experimental reasons"]
session.post_resource.return_value = post_response
session.get_resource.return_value = basic_predictor_report_data
pc = PredictorCollection(uuid.uuid4(), session)
predictor = SimpleMLPredictor.build(valid_simple_ml_predictor_data)
with pytest.warns(UserWarning) as record:
pc.register(predictor)
msg = str(record[0].message)
assert "Predictor" in msg
assert "This is a test" in msg
assert "Of experimental reasons" in msg
def test_simple_legacy_deserialization(valid_simple_ml_predictor_data):
"""Ensure that a deserialized SimplePredictor looks sane."""
predictor: SimpleMLPredictor = SimpleMLPredictor.build(
valid_simple_ml_predictor_data)
assert predictor.name == 'ML predictor'
assert predictor.description == 'Predicts z from input x and latent variable y'
assert len(predictor.inputs) == 1
assert predictor.inputs[0] == RealDescriptor("x", 0, 100, "")
assert len(predictor.outputs) == 1
assert predictor.outputs[0] == RealDescriptor("z", 0, 100, "")
assert len(predictor.latent_variables) == 1
assert predictor.latent_variables[0] == RealDescriptor("y", 0, 100, "")
assert len(predictor.training_data) == 1
assert predictor.training_data[0].table_id == UUID(
'e5c51369-8e71-4ec6-b027-1f92bdc14762')
def test_create_or_update_unique_found(predictor_collection):
# test when there is a single unique resource that exists with the listed name and update
pred = SimpleMLPredictor(
name="resource 4", #this is a unique name in the collection
description='I am updated!',
inputs=[],
outputs=[],
latent_variables=[])
#verify that the returned object is updated
returned_pred = create_or_update(predictor_collection, pred)
assert returned_pred.uid == pred.uid
assert returned_pred.name == pred.name
assert returned_pred.description == pred.description
#verify that the collection is also updated
updated_pred = [
r for r in list(predictor_collection.list()) if r.name == "resource 4"
][0]
assert updated_pred.description == "I am updated!"
def test_mark_predictor_invalid(valid_simple_ml_predictor_data, valid_predictor_report_data):
# Given
session = FakeSession()
collection = PredictorCollection(uuid.uuid4(), session)
predictor = SimpleMLPredictor.build(valid_simple_ml_predictor_data)
session.set_responses(valid_simple_ml_predictor_data, valid_predictor_report_data)
# When
predictor.archived = False
collection.update(predictor)
# Then
assert 1 == session.num_calls, session.calls
first_call = session.calls[0] # First call is the update
assert first_call.method == 'PUT'
assert first_call.path == '/projects/{}/modules/{}'.format(collection.project_id, predictor.uid)
assert not first_call.json['archived']
def test_create_or_update_none_found(predictor_collection):
# test when resource doesn't exist with listed name and check if new one is created
assert not [
r for r in list(predictor_collection.list()) if r.name == absent_name
]
pred = SimpleMLPredictor(name=absent_name,
description='',
inputs=[],
outputs=[],
latent_variables=[])
#verify that the returned object is updated
returned_pred = create_or_update(predictor_collection, pred)
assert returned_pred.uid == pred.uid
assert returned_pred.name == pred.name
assert returned_pred.description == pred.description
#verify that the collection is also updated
assert any([
r for r in list(predictor_collection.list()) if r.name == absent_name
])
