def _format_example_as_numpy_dict(example, feature_shape_dict):
result = example_coder.ExampleToNumpyDict(example)
for key, value in result.items():
shape = feature_shape_dict[key]
value = value.reshape(shape)
if not shape:
value = value.squeeze(0)
result[key] = value
return result
def test_decode_example_none_ref_count(self):
example = text_format.Parse(
'''
features {
feature {
key: 'x'
value { }
}
}
''', tf.train.Example())
before_refcount = sys.getrefcount(None)
_ = example_coder.ExampleToNumpyDict(example.SerializeToString())
after_refcount = sys.getrefcount(None)
self.assertEqual(before_refcount + 1, after_refcount)
def _ParseExample(extracts: types.Extracts, eval_config: config.EvalConfig):
"""Parses serialized tf.train.Example to create additional extracts.
Args:
extracts: PCollection containing serialized examples under tfma.INPUT_KEY.
eval_config: Eval config.
Returns:
Extracts with additional keys added for features, labels, and example
weights.
"""
features = example_coder.ExampleToNumpyDict(extracts[constants.INPUT_KEY])
extracts = copy.copy(extracts)
def add_to_extracts( # pylint: disable=invalid-name
key: Text, model_name: Text, feature_values: Any):
"""Adds features_values to extracts and feature_keys to keys_to_pop."""
# Only key by model name if multiple models.
if len(eval_config.model_specs) > 1:
if key not in extracts:
extracts[key] = {}
extracts[key][model_name] = feature_values
else:
extracts[key] = feature_values
keys_to_pop = []
for spec in eval_config.model_specs:
if spec.label_key or spec.label_keys:
keys, values = _keys_and_values(
spec.label_key or dict(spec.label_keys), features)
add_to_extracts(constants.LABELS_KEY, spec.name, values)
keys_to_pop.extend(keys)
if spec.example_weight_key or spec.example_weight_keys:
keys, values = _keys_and_values(
spec.example_weight_key or dict(spec.example_weight_keys),
features)
add_to_extracts(constants.EXAMPLE_WEIGHTS_KEY, spec.name, values)
keys_to_pop.extend(keys)
if spec.prediction_key or spec.prediction_keys:
keys, values = _keys_and_values(
spec.prediction_key or dict(spec.prediction_keys), features)
add_to_extracts(constants.PREDICTIONS_KEY, spec.name, values)
keys_to_pop.extend(keys)
for key in keys_to_pop:
if key in features:
features.pop(key)
extracts[constants.FEATURES_KEY] = features
return extracts
def test_decode_example(self, example_proto_text, decoded_example):
example = tf.train.Example()
text_format.Merge(example_proto_text, example)
self._check_decoding_results(
example_coder.ExampleToNumpyDict(example.SerializeToString()),
decoded_example)
def decode(self, serialized_example_proto: bytes) -> types.Example:
"""Decodes serialized tf.Example to tf data validation input dict."""
return example_coder.ExampleToNumpyDict(serialized_example_proto)
