def _load_eval_run(
output_path: Text
) -> Tuple[config.EvalConfig, Text, Text, Dict[Text, Text]]:
"""Returns eval config, data location, file format, and model locations."""
path = os.path.join(output_path, _EVAL_CONFIG_FILE)
if tf.io.gfile.exists(path):
with tf.io.gfile.GFile(path, 'r') as f:
pb = json_format.Parse(f.read(), config_pb2.EvalRun())
_check_version(pb.version, output_path)
return (pb.eval_config, pb.data_location, pb.file_format,
pb.model_locations)
else:
# Legacy suppport (to be removed in future).
# The previous version did not include file extension.
path = os.path.splitext(path)[0]
serialized_record = six.next(
tf.compat.v1.python_io.tf_record_iterator(path))
final_dict = pickle.loads(serialized_record)
_check_version(final_dict, output_path)
old_config = final_dict['eval_config']
slicing_specs = None
if old_config.slice_spec:
slicing_specs = [s.to_proto() for s in old_config.slice_spec]
options = config.Options()
options.compute_confidence_intervals.value = (
old_config.compute_confidence_intervals)
options.k_anonymization_count.value = old_config.k_anonymization_count
return (config.EvalConfig(slicing_specs=slicing_specs,
options=options), old_config.data_location, '', {
'': old_config.model_location
})
def _serialize_eval_run(eval_config: config.EvalConfig, data_location: Text,
file_format: Text,
model_locations: Dict[Text, Text]) -> Text:
return json_format.MessageToJson(
config_pb2.EvalRun(eval_config=eval_config,
version=tfma_version.VERSION_STRING,
data_location=data_location,
file_format=file_format,
model_locations=model_locations))
def load_eval_run(
output_path: Text,
output_file_format: Text = EVAL_CONFIG_FILE_FORMAT,
filename: Optional[Text] = None
) -> Tuple[Optional[config.EvalConfig], Text, Text, Dict[Text, Text]]:
"""Returns eval config, data location, file format, and model locations.
Args:
output_path: Directory containing config file.
output_file_format: Format of output file. Currently only 'json' is
supported.
filename: Name of output file (including extension if any).
Returns:
Tuple of (EvalConfig, data location, file format, model locations). If an
EvalConfig is not found at the given path, None will be returned.
"""
if filename is None:
filename = EVAL_CONFIG_FILE + '.' + output_file_format
path = os.path.join(output_path, filename)
if tf.io.gfile.exists(path):
with tf.io.gfile.GFile(path, 'r') as f:
pb = json_format.Parse(f.read(), config_pb2.EvalRun())
_check_version(pb.version, output_path)
return (pb.eval_config, pb.data_location, pb.file_format,
pb.model_locations)
# Legacy suppport (to be removed in future).
# The previous version did not include file extension.
path = os.path.splitext(path)[0]
if tf.io.gfile.exists(path):
serialized_record = six.next(
tf.compat.v1.python_io.tf_record_iterator(path))
final_dict = pickle.loads(serialized_record)
_check_version(final_dict, output_path)
old_config = final_dict['eval_config']
slicing_specs = None
if old_config.slice_spec:
slicing_specs = [s.to_proto() for s in old_config.slice_spec]
options = config.Options()
options.compute_confidence_intervals.value = (
old_config.compute_confidence_intervals)
options.min_slice_size.value = old_config.k_anonymization_count
return (config.EvalConfig(slicing_specs=slicing_specs,
options=options), old_config.data_location,
'', {
'': old_config.model_location
})
# No config found
return (None, '', '', {})