def test_automl_build_old(old_auto_ml_predictor_data, basic_predictor_report_data):
session = mock.Mock()
session.get_resource.return_value = basic_predictor_report_data
pc = PredictorCollection(uuid.uuid4(), session)
predictor = pc.build(old_auto_ml_predictor_data)
assert predictor.name == 'AutoML predictor'
assert predictor.description == 'Predicts z from input x'
def test_list_predictors(valid_simple_ml_predictor_data,
valid_expression_predictor_data,
basic_predictor_report_data):
# Given
session = FakeSession()
collection = PredictorCollection(uuid.uuid4(), session)
session.set_responses(
{
'entries':
[valid_simple_ml_predictor_data, valid_expression_predictor_data],
'next':
''
}, basic_predictor_report_data, basic_predictor_report_data)
# When
predictors = list(collection.list(per_page=20))
# Then
expected_call = FakeCall(method='GET',
path='/projects/{}/modules'.format(
collection.project_id),
params={
'per_page': 20,
'module_type': 'PREDICTOR'
})
assert 3 == session.num_calls, session.calls # This is a little strange, the report is fetched eagerly
assert expected_call == session.calls[0]
assert len(predictors) == 2
def test_build(valid_simple_ml_predictor_data, basic_predictor_report_data):
session = mock.Mock()
session.get_resource.return_value = basic_predictor_report_data
pc = PredictorCollection(uuid.uuid4(), session)
predictor = pc.build(valid_simple_ml_predictor_data)
assert predictor.name == 'ML predictor'
assert predictor.description == 'Predicts z from input x and latent variable y'
def test_automl_register_old(old_auto_ml_predictor_data, basic_predictor_report_data):
session = mock.Mock()
session.post_resource.return_value = old_auto_ml_predictor_data
session.get_resource.return_value = basic_predictor_report_data
pc = PredictorCollection(uuid.uuid4(), session)
predictor = AutoMLPredictor.build(old_auto_ml_predictor_data)
registered = pc.register(predictor)
assert registered.name == 'AutoML predictor'
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 test_graph_build(valid_graph_predictor_data, basic_predictor_report_data):
session = mock.Mock()
session.get_resource.return_value = basic_predictor_report_data
pc = PredictorCollection(uuid.uuid4(), session)
predictor = pc.build(valid_graph_predictor_data)
assert predictor.name == 'Graph predictor'
assert predictor.description == 'description'
assert len(predictor.predictors) == 2
assert len(predictor.training_data) == 1
def test_unexpected_pattern():
"""Check that unexpected patterns result in a value error"""
# Given
session = FakeSession()
pc = PredictorCollection(uuid.uuid4(), session)
# Then
with pytest.raises(ValueError):
pc.auto_configure(GemTableDataSource(uuid.uuid4(), 0), "yogurt")
def test_graph_register(valid_graph_predictor_data, basic_predictor_report_data):
copy_graph_data = deepcopy(valid_graph_predictor_data)
session = mock.Mock()
session.post_resource.return_value = copy_graph_data
session.get_resource.return_value = basic_predictor_report_data
pc = PredictorCollection(uuid.uuid4(), session)
predictor = GraphPredictor.build(valid_graph_predictor_data)
registered = pc.register(predictor)
assert registered.name == 'Graph predictor'
def test_get_none():
"""Test that trying to get a predictor with uid=None results in an informative error."""
session = mock.Mock()
session.get_resource.return_value = basic_predictor_report_data
pc = PredictorCollection(uuid.uuid4(), session)
with pytest.raises(ValueError) as excinfo:
pc.get(uid=None)
assert "uid=None" in str(excinfo.value)
def test_build(valid_simple_ml_data):
session = mock.Mock()
session.get_resource.return_value = {
'id': str(uuid.uuid4()),
'status': 'VALID',
'report': {}
}
pc = PredictorCollection(uuid.uuid4(), session)
predictor = pc.build(valid_simple_ml_data)
assert predictor.name == 'ML predictor'
assert predictor.description == 'Predicts z from input x and latent variable y'
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_check_update_none():
"""Test that check-for-updates makes the expected calls, parses output for no update."""
# Given
session = FakeSession()
session.set_response({"updatable": False})
pc = PredictorCollection(uuid.uuid4(), session)
predictor_id = uuid.uuid4()
# when
update_check = pc.check_for_update(predictor_id)
# then
assert update_check is None
expected_call = FakeCall(method='GET', path='/projects/{}/predictors/{}/check-for-update'.format(pc.project_id, predictor_id))
assert session.calls[0] == expected_call
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_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_returned_predictor(valid_graph_predictor_data):
"""Check that auto_configure works on the happy path."""
# Given
session = FakeSession()
# Setup a response that includes instance instead of config
response = deepcopy(valid_graph_predictor_data)
response["instance"] = response["config"]
del response["config"]
session.set_response(response)
pc = PredictorCollection(uuid.uuid4(), session)
# When
result = pc.auto_configure(GemTableDataSource(uuid.uuid4(), 0), "PLAIN")
# Then the response is parsed in a predictor
assert result.name == valid_graph_predictor_data["display_name"]
assert isinstance(result, GraphPredictor)
# including nested predictors
assert len(result.predictors) == 2
assert isinstance(result.predictors[0], uuid.UUID)
assert isinstance(result.predictors[1], DeprecatedExpressionPredictor)
def test_check_update_some():
"""Test the update check correctly builds a module."""
# given
session = FakeSession()
desc = RealDescriptor("spam", 0, 1, "kg")
response = {
"type": "AnalyticExpression",
"name": "foo",
"description": "bar",
"expression": "2 * x",
"output": RealDescriptor("spam", 0, 1, "kg").dump(),
"aliases": {}
}
session.set_response({"updatable": True, "update": response})
pc = PredictorCollection(uuid.uuid4(), session)
predictor_id = uuid.uuid4()
# when
update_check = pc.check_for_update(predictor_id)
# then
expected = ExpressionPredictor("foo", "bar", "2 * x", desc, {})
assert update_check.dump() == expected.dump()
assert update_check.uid == predictor_id
def test_archive_and_restore(valid_label_fractions_predictor_data):
session = mock.Mock()
pc = PredictorCollection(uuid.uuid4(), session)
session.get_resource.return_value = valid_label_fractions_predictor_data
def _mock_put_resource(url, data):
"""Assume that update returns the serialized predictor data."""
return data
session.put_resource.side_effect = _mock_put_resource
archived_predictor = pc.archive(uuid.uuid4())
assert archived_predictor.archived
valid_label_fractions_predictor_data["archived"] = True
session.get_resource.return_value = valid_label_fractions_predictor_data
restored_predictor = pc.restore(uuid.uuid4())
assert not restored_predictor.archived
session.get_resource.side_effect = NotFound("")
with pytest.raises(RuntimeError):
pc.archive(uuid.uuid4())
with pytest.raises(RuntimeError):
pc.restore(uuid.uuid4())
def predictors(self) -> PredictorCollection:
"""Return a resource representing all visible predictors."""
return PredictorCollection(self.uid, self.session)
def test_delete():
pc = PredictorCollection(uuid.uuid4(), mock.Mock())
with pytest.raises(NotImplementedError):
pc.delete(uuid.uuid4())