revise_preprocessed_data.py 文件源码

def filter_and_revise_example(serialized_example, samples_metadata):
  """Filter and revise a collection of existing TensorFlow examples.

  Args:
    serialized_example: the example to be revised and/or filtered
    samples_metadata: dictionary of metadata for all samples

  Returns:
    A list containing the revised example or the empty list if the
    example should be removed from the collection.
  """
  example = tf.train.Example.FromString(serialized_example)
  sample_name = example.features.feature[
      encoder.SAMPLE_NAME_FEATURE].bytes_list.value[0]
  logging.info('Checking ' + sample_name)
  if sample_name not in samples_metadata:
    logging.info('Omitting ' + sample_name)
    return []

  revised_features = {}
  # Initialize with current example features.
  revised_features.update(example.features.feature)
  # Overwrite metadata features.
  revised_features.update(
      metadata_encoder.metadata_to_ancestry_features(
          samples_metadata[sample_name]))
  return [
      tf.train.Example(features=tf.train.Features(feature=revised_features))
  ]