def _generator_fn(self, batch_size):
while True:
features = tensorspec_utils.make_random_numpy(
self._feature_spec, batch_size, self._sequence_length)
labels = tensorspec_utils.make_random_numpy(
self._label_spec, batch_size, self._sequence_length)
yield features, labels
def _create_mock_tensors(self,
base_preprocessor,
batch_size,
mode=tf.estimator.ModeKeys.TRAIN):
np.random.seed(_RANDOM_SEED)
features = utils.make_random_numpy(
base_preprocessor.get_in_feature_specification(mode),
batch_size=batch_size)
labels = utils.make_random_numpy(
base_preprocessor.get_in_label_specification(mode),
batch_size=batch_size)
return (features, labels)
def test_stack_intratask_episodes(self):
feature_spec = TSpec()
feature_spec.image = utils.ExtendedTensorSpec(
shape=_DEFAULT_IN_IMAGE_SHAPE,
dtype=tf.uint8,
is_optional=False,
data_format='jpeg',
name='state/image')
feature_spec.action = utils.ExtendedTensorSpec(
shape=_DEFAULT_ACTION_SHAPE,
dtype=tf.float32,
is_optional=False,
name='state/action')
batch_size = 2
num_samples_in_task = 3
metaexample_spec = preprocessors.create_metaexample_spec(
feature_spec, num_samples_in_task, 'condition')
tensors = utils.make_random_numpy(metaexample_spec, batch_size)
out_tensors = preprocessors.stack_intra_task_episodes(
tensors, num_samples_in_task)
self.assertEqual(
out_tensors.image.shape,
(batch_size, num_samples_in_task) + _DEFAULT_IN_IMAGE_SHAPE)
self.assertEqual(
out_tensors.action.shape,
(batch_size, num_samples_in_task) + _DEFAULT_ACTION_SHAPE)
def test_predictor(self):
input_generator = default_input_generator.DefaultRandomInputGenerator(
batch_size=_BATCH_SIZE)
model_dir = self.create_tempdir().full_path
mock_model = mocks.MockT2RModel()
train_eval.train_eval_model(t2r_model=mock_model,
input_generator_train=input_generator,
max_train_steps=_MAX_TRAIN_STEPS,
model_dir=model_dir)
predictor = checkpoint_predictor.CheckpointPredictor(
t2r_model=mock_model, checkpoint_dir=model_dir, use_gpu=False)
with self.assertRaises(ValueError):
predictor.predict({'does_not_matter': np.zeros(1)})
self.assertEqual(predictor.model_version, -1)
self.assertEqual(predictor.global_step, -1)
self.assertTrue(predictor.restore())
self.assertGreater(predictor.model_version, 0)
self.assertEqual(predictor.global_step, 3)
ref_feature_spec = mock_model.preprocessor.get_in_feature_specification(
tf.estimator.ModeKeys.PREDICT)
tensorspec_utils.assert_equal(predictor.get_feature_specification(),
ref_feature_spec)
features = tensorspec_utils.make_random_numpy(ref_feature_spec,
batch_size=_BATCH_SIZE)
predictions = predictor.predict(features)
self.assertLen(predictions, 1)
self.assertCountEqual(sorted(predictions.keys()), ['logit'])
self.assertEqual(predictions['logit'].shape, (2, 1))
def _test_predictor(self, predictor_cls):
mock_model = mocks.MockTF2T2RModel()
# Generate a sample to evaluate
feature_spec = mock_model.preprocessor.get_in_feature_specification(
tf.compat.v1.estimator.ModeKeys.PREDICT)
sample_features = tensorspec_utils.make_random_numpy(
feature_spec, batch_size=_BATCH_SIZE)
# Generate a saved model and load it.
path = self._save_model(mock_model, sample_features)
saved_model_predictor = predictor_cls(path)
# Not restored yet.
with self.assertRaises(ValueError):
saved_model_predictor.predict(sample_features)
saved_model_predictor.restore()
# Validate evaluations are the same afterwards.
original_model_out = mock_model.inference_network_fn(
sample_features, None, tf.compat.v1.estimator.ModeKeys.PREDICT)
predictor_out = saved_model_predictor.predict(sample_features)
np.testing.assert_almost_equal(original_model_out['logits'],
predictor_out['logits'])
def test_predictor(self):
input_generator = default_input_generator.DefaultRandomInputGenerator(
batch_size=_BATCH_SIZE)
model_dir = self.create_tempdir().full_path
mock_model = mocks.MockT2RModel()
train_eval.train_eval_model(
t2r_model=mock_model,
input_generator_train=input_generator,
input_generator_eval=input_generator,
max_train_steps=_MAX_TRAIN_STEPS,
eval_steps=_MAX_EVAL_STEPS,
model_dir=model_dir,
create_exporters_fn=train_eval.create_default_exporters)
predictor = exported_savedmodel_predictor.ExportedSavedModelPredictor(
export_dir=os.path.join(model_dir, 'export',
'latest_exporter_numpy'))
with self.assertRaises(ValueError):
predictor.get_feature_specification()
with self.assertRaises(ValueError):
predictor.predict({'does_not_matter': np.zeros(1)})
with self.assertRaises(ValueError):
_ = predictor.model_version
self.assertTrue(predictor.restore())
self.assertGreater(predictor.model_version, 0)
ref_feature_spec = mock_model.preprocessor.get_in_feature_specification(
tf.estimator.ModeKeys.PREDICT)
tensorspec_utils.assert_equal(predictor.get_feature_specification(),
ref_feature_spec)
features = tensorspec_utils.make_random_numpy(ref_feature_spec,
batch_size=_BATCH_SIZE)
predictions = predictor.predict(features)
self.assertLen(predictions, 1)
self.assertEqual(predictions['logit'].shape, (2, 1))
def test_predictor_with_async_hook(self):
model_dir = self.create_tempdir().full_path
default_create_export_fn = functools.partial(
async_export_hook_builder.default_create_export_fn,
batch_sizes_for_export=_BATCH_SIZES_FOR_EXPORT)
export_dir = os.path.join(model_dir, _EXPORT_DIR)
hook_builder = async_export_hook_builder.AsyncExportHookBuilder(
export_dir=export_dir, create_export_fn=default_create_export_fn)
input_generator = default_input_generator.DefaultRandomInputGenerator(
batch_size=_BATCH_SIZE)
mock_model = mocks.MockT2RModel()
train_eval.train_eval_model(t2r_model=mock_model,
input_generator_train=input_generator,
train_hook_builders=[hook_builder],
max_train_steps=_MAX_TRAIN_STEPS,
model_dir=model_dir)
predictor = exported_savedmodel_predictor.ExportedSavedModelPredictor(
export_dir=os.path.join(model_dir, _EXPORT_DIR))
with self.assertRaises(ValueError):
predictor.get_feature_specification()
with self.assertRaises(ValueError):
predictor.predict({'does_not_matter': np.zeros(1)})
with self.assertRaises(ValueError):
_ = predictor.model_version
self.assertEqual(predictor.global_step, -1)
self.assertTrue(predictor.restore())
self.assertGreater(predictor.model_version, 0)
# NOTE: The async hook builder will export the global step.
self.assertEqual(predictor.global_step, 3)
ref_feature_spec = mock_model.preprocessor.get_in_feature_specification(
tf.estimator.ModeKeys.PREDICT)
tensorspec_utils.assert_equal(predictor.get_feature_specification(),
ref_feature_spec)
features = tensorspec_utils.make_random_numpy(ref_feature_spec,
batch_size=_BATCH_SIZE)
predictions = predictor.predict(features)
self.assertLen(predictions, 1)
self.assertCountEqual(sorted(predictions.keys()), ['logit'])
self.assertEqual(predictions['logit'].shape, (2, 1))
def preprocess(preprocessor, feature_spec, label_spec, flatten=False):
with tf.Session() as sess:
feature_placeholders = tensorspec_utils.make_placeholders(
feature_spec, batch_size=1)
label_placeholders = None
if label_spec is not None:
label_placeholders = tensorspec_utils.make_placeholders(
label_spec, batch_size=1)
# Normally we want our features and labels to be flattened.
# However we support not flattened hierarchies as well.
if flatten:
feature_placeholders = tensorspec_utils.flatten_spec_structure(
feature_placeholders)
if label_spec is not None:
label_placeholders = tensorspec_utils.flatten_spec_structure(
label_placeholders)
(features_preprocessed, labels_preprocessed) = preprocessor.preprocess(
features=feature_placeholders,
labels=label_placeholders,
mode=tf.estimator.ModeKeys.TRAIN)
# We create a mapping of {key: np.array} or a namedtuple spec structure.
np_feature_spec = tensorspec_utils.make_random_numpy(
feature_spec, batch_size=1)
if label_placeholders is not None:
np_label_spec = tensorspec_utils.make_random_numpy(
label_spec, batch_size=1)
# We create our feed dict which basically consists of
# {placeholders: np.array}.
feed_dict = tensorspec_utils.map_feed_dict(feature_placeholders,
np_feature_spec,
ignore_batch=True)
if label_placeholders is not None:
feed_dict = tensorspec_utils.map_feed_dict(label_placeholders,
np_label_spec,
feed_dict,
ignore_batch=True)
fetch_results = [features_preprocessed]
if label_placeholders is not None:
fetch_results.append(labels_preprocessed)
np_preprocessed = sess.run(
fetch_results, feed_dict=feed_dict)
np_features_preprocessed = np_preprocessed[0]
if label_placeholders is not None:
np_labels_preprocessed = np_preprocessed[1]
np_feature_spec = tensorspec_utils.flatten_spec_structure(
np_feature_spec)
if label_placeholders is not None:
np_label_spec = tensorspec_utils.flatten_spec_structure(np_label_spec)
for key, value in np_feature_spec.items():
np.testing.assert_allclose(value, np_features_preprocessed[key])
if label_placeholders is not None:
for key, value in np_label_spec.items():
np.testing.assert_allclose(value, np_labels_preprocessed[key])
