def decode(self, serialized_example, items=None):
"""Decodes the given serialized TF-SequenceExample.
Args:
serialized_example: a serialized TF-SequenceExample tensor.
items: the list of items to decode. These must be a subset of the item
keys in self._items_to_handlers. If `items` is left as None, then all
of the items in self._items_to_handlers are decoded.
Returns:
the decoded items, a list of tensor.
"""
context, feature_list = parsing_ops.parse_single_sequence_example(
serialized_example, self._keys_to_context_features,
self._keys_to_sequence_features)
# Reshape non-sparse elements just once:
for k in self._keys_to_context_features:
v = self._keys_to_context_features[k]
if isinstance(v, parsing_ops.FixedLenFeature):
context[k] = array_ops.reshape(context[k], v.shape)
if not items:
items = self._items_to_handlers.keys()
outputs = []
for item in items:
handler = self._items_to_handlers[item]
keys_to_tensors = {
key: context[key] if key in context else feature_list[key]
for key in handler.keys
}
outputs.append(handler.tensors_to_item(keys_to_tensors))
return outputs
def _parse_example(example):
ctx, seq = parsing_ops.parse_single_sequence_example(
example,
context_features=fc.make_parse_example_spec_v2(ctx_cols),
sequence_features=fc.make_parse_example_spec_v2(seq_cols))
ctx.update(seq)
return ctx
def _parse_example(example):
ctx, seq = parsing_ops.parse_single_sequence_example(
example,
context_features=fc.make_parse_example_spec_v2(ctx_cols),
sequence_features=fc.make_parse_example_spec_v2(seq_cols))
ctx.update(seq)
return ctx
def _test_parsed_sequence_example(
self, col_name, col_fn, col_arg, shape, values):
"""Helper function to check that each FeatureColumn parses correctly.
Args:
col_name: string, name to give to the feature column. Should match
the name that the column will parse out of the features dict.
col_fn: function used to create the feature column. For example,
sequence_numeric_column.
col_arg: second arg that the target feature column is expecting.
shape: the expected dense_shape of the feature after parsing into
a SparseTensor.
values: the expected values at index [0, 2, 6] of the feature
after parsing into a SparseTensor.
"""
example = _make_sequence_example()
columns = [
fc.categorical_column_with_identity('int_ctx', num_buckets=100),
fc.numeric_column('float_ctx'),
col_fn(col_name, col_arg)
]
context, seq_features = parsing_ops.parse_single_sequence_example(
example.SerializeToString(),
context_features=fc.make_parse_example_spec_v2(columns[:2]),
sequence_features=fc.make_parse_example_spec_v2(columns[2:]))
with self.cached_session() as sess:
ctx_result, seq_result = sess.run([context, seq_features])
self.assertEqual(list(seq_result[col_name].dense_shape), shape)
self.assertEqual(
list(seq_result[col_name].values[[0, 2, 6]]), values)
self.assertEqual(list(ctx_result['int_ctx'].dense_shape), [1])
self.assertEqual(ctx_result['int_ctx'].values[0], 5)
self.assertEqual(list(ctx_result['float_ctx'].shape), [1])
self.assertAlmostEqual(ctx_result['float_ctx'][0], 123.6, places=1)
def _test_parsed_sequence_example(
self, col_name, col_fn, col_arg, shape, values):
"""Helper function to check that each FeatureColumn parses correctly.
Args:
col_name: string, name to give to the feature column. Should match
the name that the column will parse out of the features dict.
col_fn: function used to create the feature column. For example,
sequence_numeric_column.
col_arg: second arg that the target feature column is expecting.
shape: the expected dense_shape of the feature after parsing into
a SparseTensor.
values: the expected values at index [0, 2, 6] of the feature
after parsing into a SparseTensor.
"""
example = _make_sequence_example()
columns = [
fc.categorical_column_with_identity('int_ctx', num_buckets=100),
fc.numeric_column('float_ctx'),
col_fn(col_name, col_arg)
]
context, seq_features = parsing_ops.parse_single_sequence_example(
example.SerializeToString(),
context_features=fc.make_parse_example_spec(columns[:2]),
sequence_features=fc.make_parse_example_spec(columns[2:]))
with self.cached_session() as sess:
ctx_result, seq_result = sess.run([context, seq_features])
self.assertEqual(list(seq_result[col_name].dense_shape), shape)
self.assertEqual(
list(seq_result[col_name].values[[0, 2, 6]]), values)
self.assertEqual(list(ctx_result['int_ctx'].dense_shape), [1])
self.assertEqual(ctx_result['int_ctx'].values[0], 5)
self.assertEqual(list(ctx_result['float_ctx'].shape), [1])
self.assertAlmostEqual(ctx_result['float_ctx'][0], 123.6, places=1)
def _test(self,
kwargs,
expected_context_values=None,
expected_feat_list_values=None,
expected_err=None):
expected_context_values = expected_context_values or {}
expected_feat_list_values = expected_feat_list_values or {}
with self.test_session() as sess:
if expected_err:
with self.assertRaisesWithPredicateMatch(expected_err[0],
expected_err[1]):
c_out, fl_out = parsing_ops.parse_single_sequence_example(**kwargs)
if c_out:
sess.run(flatten_values_tensors_or_sparse(c_out.values()))
if fl_out:
sess.run(flatten_values_tensors_or_sparse(fl_out.values()))
else:
# Returns dicts w/ Tensors and SparseTensors.
context_out, feat_list_out = parsing_ops.parse_single_sequence_example(
**kwargs)
context_result = sess.run(
flatten_values_tensors_or_sparse(context_out.values(
))) if context_out else []
feat_list_result = sess.run(
flatten_values_tensors_or_sparse(feat_list_out.values(
))) if feat_list_out else []
# Check values.
_compare_output_to_expected(self, context_out, expected_context_values,
context_result)
_compare_output_to_expected(self, feat_list_out,
expected_feat_list_values, feat_list_result)
# Check shapes; if serialized is a Tensor we need its size to
# properly check.
if "context_features" in kwargs:
for k, f in kwargs["context_features"].items():
if isinstance(f, parsing_ops.FixedLenFeature) and f.shape is not None:
self.assertEqual(
tuple(context_out[k].get_shape().as_list()), f.shape)
elif isinstance(f, parsing_ops.VarLenFeature):
self.assertEqual(
tuple(context_out[k].indices.get_shape().as_list()), (None, 1))
self.assertEqual(
tuple(context_out[k].values.get_shape().as_list()), (None,))
self.assertEqual(
tuple(context_out[k].dense_shape.get_shape().as_list()), (1,))
def _test(self,
kwargs,
expected_context_values=None,
expected_feat_list_values=None,
expected_err=None):
expected_context_values = expected_context_values or {}
expected_feat_list_values = expected_feat_list_values or {}
with self.test_session() as sess:
if expected_err:
with self.assertRaisesWithPredicateMatch(expected_err[0],
expected_err[1]):
c_out, fl_out = parsing_ops.parse_single_sequence_example(**kwargs)
if c_out:
sess.run(flatten_values_tensors_or_sparse(c_out.values()))
if fl_out:
sess.run(flatten_values_tensors_or_sparse(fl_out.values()))
else:
# Returns dicts w/ Tensors and SparseTensors.
context_out, feat_list_out = parsing_ops.parse_single_sequence_example(
**kwargs)
context_result = sess.run(
flatten_values_tensors_or_sparse(context_out.values(
))) if context_out else []
feat_list_result = sess.run(
flatten_values_tensors_or_sparse(feat_list_out.values(
))) if feat_list_out else []
# Check values.
_compare_output_to_expected(self, context_out, expected_context_values,
context_result)
_compare_output_to_expected(self, feat_list_out,
expected_feat_list_values, feat_list_result)
# Check shapes; if serialized is a Tensor we need its size to
# properly check.
if "context_features" in kwargs:
for k, f in kwargs["context_features"].items():
if isinstance(f, parsing_ops.FixedLenFeature) and f.shape is not None:
self.assertEqual(
tuple(context_out[k].get_shape().as_list()), f.shape)
elif isinstance(f, parsing_ops.VarLenFeature):
self.assertEqual(
tuple(context_out[k].indices.get_shape().as_list()), (None, 1))
self.assertEqual(
tuple(context_out[k].values.get_shape().as_list()), (None,))
self.assertEqual(
tuple(context_out[k].dense_shape.get_shape().as_list()), (1,))
def parse_feature_columns_from_sequence_examples(
serialized,
context_feature_columns,
sequence_feature_columns,
name=None,
example_name=None):
"""Parses tf.SequenceExamples to extract tensors for given `FeatureColumn`s.
Args:
serialized: A scalar (0-D Tensor) of type string, a single serialized
`SequenceExample` proto.
context_feature_columns: An iterable containing the feature columns for
context features. All items should be instances of classes derived from
`_FeatureColumn`. Can be `None`.
sequence_feature_columns: An iterable containing the feature columns for
sequence features. All items should be instances of classes derived from
`_FeatureColumn`. Can be `None`.
name: A name for this operation (optional).
example_name: A scalar (0-D Tensor) of type string (optional), the names of
the serialized proto.
Returns:
A tuple consisting of (context_features, sequence_features)
* context_features: a dict mapping `FeatureColumns` from
`context_feature_columns` to their parsed `Tensors`/`SparseTensor`s.
* sequence_features: a dict mapping `FeatureColumns` from
`sequence_feature_columns` to their parsed `Tensors`/`SparseTensor`s.
"""
# Sequence example parsing requires a single (scalar) example.
try:
serialized = array_ops.reshape(serialized, [])
except ValueError as e:
raise ValueError(
'serialized must contain as single sequence example. Batching must be '
'done after parsing for sequence examples. Error: {}'.format(e))
if context_feature_columns is None:
context_feature_columns = []
if sequence_feature_columns is None:
sequence_feature_columns = []
check_feature_columns(context_feature_columns)
context_feature_spec = fc.create_feature_spec_for_parsing(
context_feature_columns)
check_feature_columns(sequence_feature_columns)
sequence_feature_spec = fc._create_sequence_feature_spec_for_parsing( # pylint: disable=protected-access
sequence_feature_columns, allow_missing_by_default=False)
return parsing_ops.parse_single_sequence_example(serialized,
context_feature_spec,
sequence_feature_spec,
example_name,
name)
def serialize_map(replay_example_str):
"""Parse each example string to `tf.Tensor`."""
try:
assert_op = control_flow_ops.Assert(replay_example_str != "",
[replay_example_str])
with ops.control_dependencies([assert_op]):
_, replay = parsing_ops.parse_single_sequence_example(
replay_example_str, sequence_features=replay_features)
except errors_impl.InvalidArgumentError:
raise errors_impl.OutOfRangeError()
return convert_and_fix_dtypes(replay)
def main():
reader = tf.TFRecordReader
data_sources = ["traineh.tfrecord"]
_, data = parallel_reader.parallel_read(
data_sources,
reader_class=reader,
num_epochs=1,
num_readers=1,
shuffle=False,
capacity=256,
min_after_dequeue=1)
context_features, sequence_features = parsing_ops.parse_single_sequence_example(data, context_features={
'video_id': tf.VarLenFeature(tf.string),
'labels': tf.VarLenFeature(tf.int64),
}, sequence_features={
'inc3': tf.FixedLenSequenceFeature(1, tf.string)
}, example_name="")
with tf.Session() as sess:
sess.run(tf.initialize_local_variables())
sess.run(tf.initialize_all_variables())
coord = tf.train.Coordinator()
threads = tf.train.start_queue_runners(sess=sess, coord=coord)
try:
while not coord.should_stop():
meta = sess.run(context_features)
vid = meta['video_id'].values[0]
labels = meta['labels'].values
inc3_fea = sess.run(sequence_features)['inc3']
frame_feas = []
for r in inc3_fea:
v = np.fromstring(r[0], dtype=np.uint8)
frame_feas.append(v[None, :])
frame_feas = np.vstack(frame_feas)
print(vid, labels)
print(frame_feas.shape)
# Do something here
except tf.errors.OutOfRangeError:
print('Finished extracting.')
finally:
coord.request_stop()
coord.join(threads)
def _convert_Youtube8M_tfrecord_to_numpy(tfrecord_filename):
'''
Function:
_convert_Youtube8M_tfrecord_to_numpy
i.e. parse each data_component according to example_prototxt
Input:
<string> tfrecord_filename
Output:
<dictionary> parsed_data
'''
reader = tf.TFRecordReader
_, data = parallel_reader.parallel_read(data_sources=tfrecord_filename,
reader_class=reader,
num_epochs=1,
num_readers=1,
shuffle=False,
capacity=256,
min_after_dequeue=1)
# build-up fileQueue and exampleQueue for tfrecords.file...
context_feat, seq_feat = parsing_ops.parse_single_sequence_example(
data,
context_features={
'video_id': tf.VarLenFeature(tf.string),
'labels': tf.VarLenFeature(tf.int64)
},
sequence_features={
'rgb': tf.FixedLenSequenceFeature([], tf.string),
'audio': tf.FixedLenSequenceFeature([], tf.string)
},
example_name=" ")
# standard framework for example parsing...
with tf.Session() as sess:
#--- initialize variables in tensorflow session ---#
sess.run(tf.local_variables_initializer())
sess.run(tf.global_variables_initializer())
#--- start-up coordinator to manage the QueueRunner threads ---#
coord = tf.train.Coordinator()
threads = tf.train.start_queue_runners(sess=sess, coord=coord)
#--- training operations ---#
try:
total_rgb_feat = []
total_audio_feat = []
total_label = []
while not coord.should_stop():
video_context, video_features = sess.run(
(context_feat, seq_feat))
#--- extract 'video_id' and 'labels' from context features ---#
video_id = video_context['video_id'].values[0]
labels = video_context['labels'].values
#--- one-hot vector for labels ---#
labels = sess.run(
tf.sparse_to_dense(labels, (4716, ),
1,
validate_indices=False))
#--- extract 'rgb' and 'audio' features from video features ---#
hex_rgb_feat = video_features['rgb']
hex_audio_feat = video_features['audio']
rgb_feat = []
audio_feat = []
#--- convert hex data i.e. hex_rgb_feat to numpy.uint8 format ---#
for ii in range(len(hex_rgb_feat)):
single_rgb_feat = np.fromstring(hex_rgb_feat[ii],
dtype=np.uint8)
single_audio_feat = np.fromstring(hex_audio_feat[ii],
dtype=np.uint8)
rgb_feat.append(single_rgb_feat)
audio_feat.append(single_audio_feat)
#--- reshape e.g. [[1,2], [3,4]] -> [1,2; 3,4]
rgb_feat = np.vstack(rgb_feat)
audio_feat = np.vstack(audio_feat)
#--- dequantize the rgb and audio features... ---#
rgb_feat = _dequantize(rgb_feat, 2, -2)
audio_feat = _dequantize(audio_feat, 2, -2)
#--- padding or crop to fixed nframe=300... ---#
rgb_feat = _frame_padding(input_feat=rgb_feat,
padding_value=0,
target_nframe=300)
audio_feat = _frame_padding(input_feat=audio_feat,
padding_value=0,
target_nframe=300)
total_rgb_feat.append(rgb_feat)
total_audio_feat.append(audio_feat)
total_label.append(labels)
except tf.errors.OutOfRangeError:
print('!All video features have been exported...')
finally:
coord.request_stop()
coord.join(threads=threads)
return total_rgb_feat, total_audio_feat, total_label
sess.close()
