def package_simple_addition_model(self, do_fail=False):
with TemporaryDirectory() as test_dir:
neuropod_path = os.path.join(test_dir, "test_neuropod")
model_code_dir = os.path.join(test_dir, "model_code")
os.makedirs(model_code_dir)
with open(os.path.join(model_code_dir, "addition_model.py"),
"w") as f:
f.write(ADDITION_MODEL_SOURCE)
# `create_python_neuropod` runs inference with the test data immediately
# after creating the neuropod. Raises a ValueError if the model output
# does not match the expected output.
create_python_neuropod(
neuropod_path=neuropod_path,
model_name="addition_model",
data_paths=[],
code_path_spec=[{
"python_root":
model_code_dir,
"dirs_to_package": [
"" # Package everything in the python_root
],
}],
entrypoint_package="addition_model",
entrypoint="get_model",
# Get the input/output spec along with test data
**get_addition_model_spec(do_fail=do_fail))
# Run some additional checks
check_addition_model(neuropod_path)
def test_noncontiguous_array(self):
x = np.arange(16).astype(np.int64).reshape(4, 4)
with TemporaryDirectory() as test_dir:
neuropod_path = os.path.join(test_dir, "test_neuropod")
model_code_dir = os.path.join(test_dir, "model_code")
os.makedirs(model_code_dir)
with open(os.path.join(model_code_dir, "splitter_model.py"),
"w") as f:
f.write(NONCONTIGUOUS_MODEL_SOURCE)
create_python_neuropod(
neuropod_path=neuropod_path,
model_name="splitter",
data_paths=[],
code_path_spec=[{
"python_root":
model_code_dir,
"dirs_to_package": [
"" # Package everything in the python_root
],
}],
entrypoint_package="splitter_model",
entrypoint="get_model",
input_spec=[{
"name": "x",
"dtype": "int64",
"shape": (4, 4)
}],
output_spec=[
{
"name": "x1",
"dtype": "int64",
"shape": (4, 2)
},
{
"name": "x2",
"dtype": "int64",
"shape": (4, 2)
},
],
test_input_data={"x": x},
test_expected_out={
"x1": x[:, :2],
"x2": x[:, 2:]
},
)
def package_sklearn_model(self, test_dir):
neuropod_path = os.path.join(test_dir, "test_neuropod")
model_code_dir = os.path.join(test_dir, "model_code")
os.makedirs(model_code_dir)
with open(os.path.join(model_code_dir, "sklearn_model.py"), "w") as f:
f.write(SKLEARN_MODEL_SOURCE)
create_python_neuropod(
neuropod_path=neuropod_path,
model_name="sklearn_model",
data_paths=[],
code_path_spec=[{
"python_root": model_code_dir,
"dirs_to_package":
[""], # Package everything in the python_root
}],
entrypoint_package="sklearn_model",
entrypoint="get_model",
input_spec=[{
"name": "x",
"dtype": "float64",
"shape": ("batch_size", 2)
}],
output_spec=[{
"name": "out",
"dtype": "float64",
"shape": ("batch_size", )
}],
test_input_data={"x": np.array([[3, 5]], dtype=np.float64)},
test_expected_out={"out": np.array([16], dtype=np.float64)},
# We define requirements this way because this test runs on python 2.7 - 3.8
# but there isn't a single version of scikit-learn that works on all of them
requirements="""
# Requirements for this model
scikit-learn=={}
""".format("0.20.0" if sys.version_info.major == 2 else "0.22.0"),
)
return neuropod_path
def package_dummy_model(self, test_dir, target):
neuropod_path = os.path.join(test_dir, "test_neuropod")
model_code_dir = os.path.join(test_dir, "model_code")
os.makedirs(model_code_dir)
with open(os.path.join(model_code_dir, "dummy_model.py"), "w") as f:
f.write(DUMMY_MODEL_SOURCE.format(target))
# `create_python_neuropod` runs inference with the test data immediately
# after creating the neuropod. Raises a ValueError if the model output
# does not match the expected output.
create_python_neuropod(
neuropod_path=neuropod_path,
model_name="dummy_model",
data_paths=[],
code_path_spec=[{
"python_root": model_code_dir,
"dirs_to_package":
[""], # Package everything in the python_root
}],
entrypoint_package="dummy_model",
entrypoint="get_model",
input_spec=[{
"name": "x",
"dtype": "float32",
"shape": (1, )
}],
output_spec=[{
"name": "out",
"dtype": "float32",
"shape": (1, )
}],
test_input_data={"x": np.array([0.0], dtype=np.float32)},
test_expected_out={"out": np.array([target], dtype=np.float32)},
)
return neuropod_path