def test_generic_serde():
gen = SizedTestType(10)
payload = serialize(gen)
assert payload == {'size': 10}
new_gen = deserialize(payload, SizedTestType)
assert issubclass(new_gen, SizedTestType)
assert new_gen.size == 10
obj = AClass([1 for _ in range(10)])
assert serialize(obj, gen) == serialize(obj, new_gen)
def test_hierarchy_serde():
gen = ChildGenericType(10)
payload = serialize(gen)
assert payload == {'size': 10, 'type': 'child'}
new_gen = deserialize(payload, AbstractType)
assert issubclass(new_gen, ChildGenericType)
assert new_gen.size == 10
obj = AClass([1 for _ in range(10)])
assert serialize(obj, gen) == serialize(obj, new_gen)
def __call__(self, *args, **kwargs):
if args and kwargs:
raise ValueError(
'Parameters should be passed either in positional or in keyword fashion, not both'
)
if len(args) > len(self.method.args) or len(kwargs) > len(
self.method.args):
raise ValueError(
f'Too much parameters given, expected: {len(self.method.args)}'
)
data = {}
for i, arg in enumerate(self.method.args):
obj = None
if len(args) > i:
obj = args[i]
if arg.name in kwargs:
obj = kwargs[arg.name]
if obj is None:
raise ValueError(
f'Parameter with name "{arg.name}" (position {i}) should be passed'
)
data[arg.name] = serialize(obj, arg.type)
logger.debug('Calling server method "%s", args: %s ...',
self.method.name, data)
out = self.call_method(self.method.name, data)
logger.debug('Server call returned %s', out)
return deserialize(out, self.method.out_type)
def serde_and_compare(obj,
obj_type=None,
true_payload=None,
check_payload=True):
if obj_type is None:
obj_type = type(obj)
check_subtype = False
check_instance = True
else:
check_subtype = not issubclass(obj_type, Serializer)
check_instance = False
payload = pyjackson.serialize(obj, obj_type)
if true_payload is not None:
if check_payload:
assert true_payload == payload
payload = true_payload
new_obj = pyjackson.deserialize(payload, obj_type)
if check_subtype:
assert issubclass(type(new_obj),
obj_type), '{} type must be subtype of {}'.format(
new_obj, obj_type)
elif check_instance:
assert isinstance(new_obj, obj_type)
assert obj == new_obj
def main():
ebnt = ebonite.Ebonite.local(clear=True)
data, target = get_data()
# we want easy way to transform anything to datasets, so its either this or ebonite.create_dataset (same for metrics)
# for now there is no difference, however if we want manage datasets with meta and art repos, we use client
# or create with ebonite.create_... and then push with ebnt.push_... like for models
# dataset = ebnt.create_dataset(data, target)
# here we postpone setting task input and output types for easy task creation
task = ebnt.get_or_create_task('my_project', 'regression_is_my_profession')
task.add_metric('auc', roc_auc_score)
task.add_metric('custom', my_custom_metric)
task.add_evaluation('train', data, target, ['auc', 'custom'])
pprint(task.evaluation_sets)
pprint(task.datasets)
pprint(task.metrics)
# omit providing dataset as we already have it in task
mc = task.create_and_push_model(constant, data, model_name='constant')
mt = task.create_and_push_model(truth, data, model_name='truth')
pprint(mc.wrapper.methods)
pprint(mt.wrapper.methods)
# maybe save result to models? also need different ways to evaluate "not all"
result = task.evaluate_all()
print(result)
ebnt._bind(task)
task.save()
pprint(serialize(task))
def test_unordered_columns(df_type, data):
data_rev = data[list(reversed(data.columns))]
obj = serialize(data_rev, df_type)
data2 = deserialize(obj, df_type)
assert data.equals(data2), f'{data} \n!=\n{data2}'
assert data2 is not data
def test_dataframe_type(df_type):
assert df_type.requirements.modules == ['pandas']
data = pd.DataFrame([{'a': 1, 'b': 1}, {'a': 2, 'b': 2}])
obj = serialize(data, df_type)
data2 = deserialize(obj, df_type)
assert data.equals(data2)
def test_df_type(df_type_fx, request):
df_type = request.getfixturevalue(df_type_fx)
assert issubclass(df_type, DataFrameType)
obj = serialize(df_type)
new_df_type = deserialize(obj, DatasetType)
assert df_type == new_df_type
def test_dict_with_list_dataset_type():
data = {'a': ['b']}
dt = DatasetAnalyzer.analyze(data)
assert dt == DictDatasetType(
{'a': TupleLikeListDatasetType([PrimitiveDatasetType('str')])})
assert serialize(data, dt) == data
assert deserialize(data, dt) == data
with pytest.raises(DeserializationError):
deserialize('', dt)
with pytest.raises(SerializationError):
serialize('', dt)
payload = serialize(dt)
assert payload == {
'type': 'dict',
'item_types': {
'a': {
'type': 'tuple_like_list',
'items': [{
'type': 'primitive',
'ptype': 'str'
}]
}
}
}
payload = serialize(DTHolder(dt))
assert payload == {
'dt': {
'type': 'dict',
'item_types': {
'a': {
'type': 'tuple_like_list',
'items': [{
'type': 'primitive',
'ptype': 'str'
}]
}
}
}
}
def test_set_hint():
@make_string
class CClass(Comparable):
def __init__(self, value: Set[str]):
self.value = value
value = CClass({'a', 'b'})
serde_and_compare(value)
assert serialize(value) in [{'value': ['a', 'b']}, {'value': ['b', 'a']}]
def test_unsized(times):
real_data = [[1, 2] for _ in range(times)]
array = MockNumpyNdarray(real_data)
container = MultidimUnsizedArrayContainer(array)
ser = serialize(container)
assert real_data == ser['arr']
new_container = deserialize(ser, MultidimUnsizedArrayContainer)
assert new_container == container
def test_dataframe_type():
data = pd.DataFrame([{'a': 1, 'b': 1}, {'a': 2, 'b': 2}])
df_type = DataFrameType(['a', 'b'])
obj = serialize(data, df_type)
data2 = deserialize(obj, df_type)
assert data.equals(data2)
def test_dataframe_type(df_type, data):
assert df_type.requirements.modules == ['pandas']
obj = serialize(data, df_type)
payload = json.dumps(obj)
loaded = json.loads(payload)
data2 = deserialize(loaded, df_type)
assert data.equals(data2)
def test_inner_hierarchy_serde():
holder = GenericTypeHolder(ChildGenericType(15))
payload = serialize(holder)
assert payload == {'gen_type': {'type': 'child', 'size': 15}}
new_holder = deserialize(payload, GenericTypeHolder)
assert holder == new_holder
def test_all(data2):
df_type = DatasetAnalyzer.analyze(data2)
obj = serialize(data2, df_type)
payload = json.dumps(obj)
loaded = json.loads(payload)
data = deserialize(loaded, df_type)
assert data2.equals(data)
assert data2 is not data
def wrapper_meta(self) -> dict:
"""
:return: pyjackson representation of :class:`~ebonite.core.objects.wrapper.ModelWrapper` for this model: e.g.,
this provides possibility to move a model between repositories without its dependencies being installed
"""
if self._wrapper_meta is None:
if self._wrapper is None:
raise ValueError("Either 'wrapper' or 'wrapper_meta' should be provided")
self._wrapper_meta = serialize(self._wrapper)
return self._wrapper_meta
def test_all(df):
df_type = DatasetAnalyzer.analyze(df)
obj = serialize(df, df_type)
payload = json.dumps(obj)
loaded = json.loads(payload)
data = deserialize(loaded, df_type)
assert df is not data
pandas_assert(data, df)
def test_multidim():
real_data = [[1, 2, 3], [4, 5, 6]]
ext_type = MockNumpyNdarray(real_data)
c = MultidimArrayContainer(ext_type)
ser = serialize(c)
assert real_data == ser['arr']
deser = deserialize(ser, MultidimArrayContainer)
assert deser == c
def test_sized():
real_data = [1, 2, 3]
ext_type = MockNumpyNdarray(real_data)
c = SizedArrayContainer(ext_type)
ser = serialize(c)
assert real_data == ser['arr']
deser = deserialize(ser, SizedArrayContainer)
assert deser == c
def save(self, path: str = None):
path = path or SSCIConf.CONFIG_PATH
try:
with open(path, "r") as f:
data = yaml.safe_load(f)
except FileNotFoundError:
data = {"ssci": {}}
data["ssci"]["deploy"] = serialize(self)
with open(path, "w") as f:
yaml.safe_dump(data, f)
def test_set():
value = {AClass('a'), AClass('b')}
serde_and_compare(value, Set[AClass])
assert serialize(value, Set[AClass]) in [[{
'value': 'a'
}, {
'value': 'b'
}], [{
'value': 'b'
}, {
'value': 'a'
}]]
def test_type_hierarchy__type_import():
payload = {
'type': 'tests.not_imported_directly.ChildClass',
'field': 'aaa'
}
obj = deserialize(payload, RootClass)
assert isinstance(obj, RootClass)
assert obj.__class__.__name__ == 'ChildClass'
assert obj.field == 'aaa'
new_payload = serialize(obj)
assert new_payload == payload
def test_list_dataset_type(ldt):
assert ldt == ListDatasetType(PrimitiveDatasetType('int'), 2)
payload = serialize(ldt)
assert payload == {
'type': 'list',
'dtype': {
'type': 'primitive',
'ptype': 'int'
},
'size': 2
}
def test_pipeline__load(meta, model, task_saved_art):
task_saved_art.push_model(model)
p = model.as_pipeline('predict')
task_saved_art.add_pipeline(p)
p = deserialize(serialize(meta.get_pipeline_by_id(p.id)), Pipeline)
assert p is not None
assert len(p.models) == 0
p.bind_meta_repo(meta)
p.load()
assert len(p.models) == 1
assert model.name in p.models
assert p.models[model.name] == model
def test_dict_dataset_type(ddt):
assert ddt == DictDatasetType({'a': PrimitiveDatasetType('int')})
payload = serialize(ddt)
assert payload == {
'type': 'dict',
'item_types': {
"a": {
'type': 'primitive',
'ptype': 'int'
}
}
}
def test_inner_list_dataset_type(ldt):
dth = DTHolder(ldt)
payload = serialize(dth)
assert payload == {
'dt': {
'type': 'list',
'dtype': {
'type': 'primitive',
'ptype': 'int'
},
'size': 2
}
}
def test_inner_dict_dataset_type(ddt):
dth = DTHolder(ddt)
payload = serialize(dth)
assert payload == {
'dt': {
'type': 'dict',
'item_types': {
"a": {
'type': 'primitive',
'ptype': 'int'
}
}
}
}
def test_inner_tuple_like_list_dataset_type(tlldt):
dth = DTHolder(tlldt)
payload = serialize(dth)
assert payload == {
'dt': {
'type':
'tuple_like_list',
'items': [{
'type': 'primitive',
'ptype': 'str'
}, {
'type': 'primitive',
'ptype': 'int'
}]
}
}
def test_tuple_like_list_dataset_type(tlldt):
assert tlldt == TupleLikeListDatasetType(
[PrimitiveDatasetType('str'),
PrimitiveDatasetType('int')])
payload = serialize(tlldt)
assert payload == {
'type':
'tuple_like_list',
'items': [{
'type': 'primitive',
'ptype': 'str'
}, {
'type': 'primitive',
'ptype': 'int'
}]
}
def test_datetime():
data = pd.DataFrame([{
'a': 1,
'b': datetime.now()
}, {
'a': 2,
'b': datetime.now()
}])
df_type = DatasetAnalyzer.analyze(data)
assert issubclass(df_type, DataFrameType)
obj = serialize(data, df_type)
payload = json.dumps(obj)
loaded = json.loads(payload)
data2 = deserialize(loaded, df_type)
assert data.equals(data2)
assert data2 is not data
