def test_yaml_save_load_roundtrip(extension):
data = {"some": ["data"]}
with NamedTemporaryFile() as f:
path = Path(f.name).with_suffix(extension)
save_to_yaml(data, path)
f.flush()
data_deserialized = load_yaml(path)
assert data == data_deserialized
def to_yaml(self, path: Pathlike):
data = self.to_dict()
# Some feature extractors might have a "device" field:
# to make sure they get nicely serialized to YAML, we will convert
# the torch.device object to its string type.
# Note: we don't store the device ID (e.g. we change "cuda:1" to "cuda")
# so that the config remains valid even if we use it in a separate run
# on a different device.
if "device" in data and isinstance(data["device"], torch.device):
data["device"] = data["device"].type
save_to_yaml(data, path=path)
def to_yaml(self, path: Pathlike):
data = asdict(self.config)
data['feature_type'] = self.name # Insert the typename for config readability
save_to_yaml(data, path=path)
