def double_model():
def f1(a):
return a + a
model = Model.create(f1, 'a', '1')
model._id = 1
return model
def len_model():
def f2(a):
return len(a)
model = Model.create(f2, 'a', '2')
model._id = 2
return model
def test_create_model(sklearn_model_obj, pandas_data):
model = Model.create(sklearn_model_obj, pandas_data)
assert model is not None
assert isinstance(model.wrapper, SklearnModelWrapper)
input_meta, output_meta = model.wrapper.method_signature('predict')
assert input_meta.columns == list(pandas_data)
assert output_meta.real_type == np.ndarray
assert {'numpy', 'sklearn', 'pandas'}.issubset(model.requirements.modules)
def test_create_model_with_custom_wrapper(sklearn_model_obj, pandas_data):
wrapper = SklearnModelWrapper().bind_model(sklearn_model_obj, input_data=pandas_data)
model = Model.create(sklearn_model_obj, pandas_data, custom_wrapper=wrapper)
assert model is not None
assert model.wrapper is wrapper
input_meta, output_meta = model.wrapper.method_signature('predict')
assert input_meta.columns == list(pandas_data)
assert output_meta.real_type == np.ndarray
assert {'numpy', 'sklearn', 'pandas'}.issubset(model.requirements.modules)
def test_create_model(sklearn_model_obj, pandas_data):
model = Model.create(sklearn_model_obj, pandas_data)
assert model is not None
assert isinstance(model.wrapper, SklearnModelWrapper)
assert model.input_meta.columns == list(pandas_data)
# assert model.input_meta. == data.values
assert model.output_meta.real_type == np.ndarray
assert {'numpy', 'sklearn', 'pandas'}.issubset(model.requirements.modules)
def test_create_model_with_custom_wrapper(sklearn_model_obj, pandas_data):
wrapper = SklearnModelWrapper().bind_model(sklearn_model_obj)
model = Model.create(sklearn_model_obj,
pandas_data,
custom_wrapper=wrapper)
assert model is not None
assert isinstance(model.wrapper, SklearnModelWrapper)
assert model.input_meta.columns == list(pandas_data)
assert model.output_meta.real_type == np.ndarray
assert {'numpy', 'sklearn', 'pandas'}.issubset(model.requirements.modules)
def test_create_model_with_additional_artifact(artifact, sklearn_model_obj, pandas_data, artifact_repository):
model = Model.create(sklearn_model_obj, pandas_data, additional_artifacts=artifact)
assert model is not None
model._id = 'test_model'
artifact_repository.push_model_artifacts(model)
assert len(model.artifact_req_persisted.bytes_dict()) == 4
model_payloads = model.artifact_req_persisted.bytes_dict()
for name, payload in artifact.bytes_dict().items():
assert name in model_payloads
assert model_payloads[name] == payload
def test_create_model_with_custom_requirements(sklearn_model_obj, pandas_data):
requirements = Requirements([
InstallableRequirement('dumb', '0.4.1'),
InstallableRequirement('art', '4.0')
])
model = Model.create(sklearn_model_obj,
pandas_data,
custom_requirements=Requirements([Requirement()]))
assert model is not None
assert all(req in [r.module for r in requirements.installable]
for req in model.requirements.installable)
def create_model(model_object,
input_data,
model_name: str = None,
params: Dict[str, Any] = None,
description: str = None) -> Model:
"""
Creates Model instance from arbitrary model objects and sample of input data
:param model_object: model object (function, sklearn model, tensorflow output tensor list etc)
:param input_data: sample of input data (numpy array, pandas dataframe, feed dict etc)
:param model_name: name for model in database, if not provided will be autogenerated
:param params: dict with arbitrary parameters. Must be json-serializable
:param description: text description of this model
:return: :class:`~ebonite.core.objects.core.Model` instance
"""
return Model.create(model_object, input_data, model_name, params,
description)
def test_multimodel_buildable(metadata_repo):
# Dunno why, but it only worked w/o fixtures
proj = Project('proj')
task = Task('Test Task')
mdl = Model.create(lambda data: data, 'input', 'test_model')
proj = metadata_repo.create_project(proj)
task.project = proj
task = metadata_repo.create_task(task)
mdl.task = task
mdl = metadata_repo.create_model(mdl)
with pytest.raises(ValueError):
MultiModelBuildable([], server_type=FlaskServer.type)
assert mdl.has_meta_repo
mm_buildable = MultiModelBuildable([mdl], server_type=FlaskServer.type)
assert mm_buildable.task.name == 'Test Task'
assert mm_buildable.get_provider().get_python_version(
) == platform.python_version()
assert len(mm_buildable.models) == 1
def model():
model = Model.create(func, "kek", "Test Model")
return model
def unpersisted_model(sklearn_model_obj, pandas_data):
model = Model.create(sklearn_model_obj, pandas_data)
model._id = 'test_model'
assert model._persisted_artifacts is None
assert model._unpersisted_artifacts is not None
return model
def create_test_model(name):
model = Model.create(func, "kek", name)
return model
def test_create_model_with_custom_input_meta(sklearn_model_obj, pandas_data):
model = Model.create(sklearn_model_obj,
pandas_data,
custom_input_meta=DataFrameType(['kek1', 'kek2']))
assert model is not None
assert issubclass(model.input_meta, DataFrameType)
def test_base_author(sklearn_model_obj, pandas_data, username):
model = Model.create(sklearn_model_obj, pandas_data)
assert model is not None
assert model.author == username
def created_model(sklearn_model_obj, pandas_data):
return Model.create(sklearn_model_obj, pandas_data)
def model():
mdl = Model.create(lambda data: data, 'input', 'test_model')
mdl._id = 'test_model_id'
return mdl
