Attend the Women in ML Symposium on December 7 Register now

Generating big datasets with Apache Beam

Stay organized with collections Save and categorize content based on your preferences.

Some datasets are too big to be processed on a single machine. tfds supports generating data across many machines by using Apache Beam.

This doc has two sections:

  • For user who want to generate an existing Beam dataset
  • For developers who want to create a new Beam dataset

Generating a Beam dataset

Below are different examples of generating a Beam dataset, both on the cloud or locally.

On Google Cloud Dataflow

To run the pipeline using Google Cloud Dataflow and take advantage of distributed computation, first follow the Quickstart instructions.

Once your environment is set up, you can run the tfds build CLI using a data directory on GCS and specifying the required options for the --beam_pipeline_options flag.

To make it easier to launch the script, it's helpful to define the following variables using the actual values for your GCP/GCS setup and the dataset you want to generate:


You will then need to create a file to tell Dataflow to install tfds on the workers:

echo "tensorflow_datasets[$DATASET_NAME]" > /tmp/beam_requirements.txt

If you're using tfds-nightly, make sure to to echo from tfds-nightly in case the dataset has been updated since the last release.

echo "tfds-nightly[$DATASET_NAME]" > /tmp/beam_requirements.txt

Finally, you can launch the job using the command below:

  --data_dir=$GCS_BUCKET/tensorflow_datasets \


To run your script locally using the default Apache Beam runner, the command is the same as for other datasets:

tfds build my_dataset

With a custom script

To generate the dataset on Beam, the API is the same as for other datasets. You can customize the beam.Pipeline using the beam_options (and beam_runner) arguments of DownloadConfig.

# If you are running on Dataflow, Spark,..., you may have to set-up runtime
# flags. Otherwise, you can leave flags empty [].
flags = ['--runner=DataflowRunner', '--project=<project-name>', ...]

# `beam_options` (and `beam_runner`) will be forwarded to `beam.Pipeline`
dl_config =
data_dir = 'gs://my-gcs-bucket/tensorflow_datasets'
builder = tfds.builder('wikipedia/20190301.en', data_dir=data_dir)

Implementing a Beam dataset


In order to write Apache Beam datasets, you should be familiar with the following concepts:


If you are familiar with the dataset creation guide, adding a Beam dataset only requires to modify the _generate_examples function. The function should returns a beam object, rather than a generator:

Non-beam dataset:

def _generate_examples(self, path):
  for f in path.iterdir():
    yield _process_example(f)

Beam dataset:

def _generate_examples(self, path):
  return (
      | beam.Map(_process_example)

All the rest can be 100% identical, including tests.

Some additional considerations:

  • Use tfds.core.lazy_imports to import Apache Beam. By using a lazy dependency, users can still read the dataset after it has been generated without having to install Beam.
  • Be careful with Python closures. When running the pipeline, the beam.Map and beam.DoFn functions are serialized using pickle and sent to all workers. Do not use mutable objects inside a beam.PTransform if the state has to be shared across workers.
  • Due to the way tfds.core.DatasetBuilder is serialized with pickle, mutating tfds.core.DatasetBuilder during data creation will be ignored on the workers (e.g. it's not possible to set['offset'] = 123 in _split_generators and access it from the workers like beam.Map(lambda x: x +['offset']))
  • If you need to share some pipeline steps between the splits, you can add add an extra pipeline: beam.Pipeline kwarg to _split_generator and control the full generation pipeline. See _generate_examples documentation of tfds.core.GeneratorBasedBuilder.


Here is an example of a Beam dataset.

class DummyBeamDataset(tfds.core.GeneratorBasedBuilder):

  VERSION = tfds.core.Version('1.0.0')

  def _info(self):
    return self.dataset_info_from_configs(
            'image': tfds.features.Image(shape=(16, 16, 1)),
            'label': tfds.features.ClassLabel(names=['dog', 'cat']),

  def _split_generators(self, dl_manager):
    return {
        'train': self._generate_examples(file_dir='path/to/train_data/'),
        'test': self._generate_examples(file_dir='path/to/test_data/'),

  def _generate_examples(self, file_dir: str):
    """Generate examples as dicts."""
    beam = tfds.core.lazy_imports.apache_beam

    def _process_example(filename):
      # Use filename as key
      return filename, {
          'image': os.path.join(file_dir, filename),
          'label': filename.split('.')[1],  # Extract label: ""

    return (
        | beam.Map(_process_example)

Running your pipeline

To run the pipeline, have a look at the above section.

tfds build my_dataset --register_checksums

Pipeline using TFDS as input

If you want to create a beam pipeline which takes a TFDS dataset as source, you can use the tfds.beam.ReadFromTFDS:

builder = tfds.builder('my_dataset')

_ = (
    | tfds.beam.ReadFromTFDS(builder, split='train')
    | beam.Map(tfds.as_numpy)
    | ...

It will process each shard of the dataset in parallel.