TensorFlow I/O

TensorFlow I/O is a collection of file systems and file formats that are not available in TensorFlow’s built-in support.

At the moment TensorFlow I/O supports the following data sources:

• tensorflow_io.ignite: Data source for Apache Ignite and Ignite File System (IGFS). Overview and usage guide here.
• tensorflow_io.kafka: Apache Kafka stream-processing support.
• tensorflow_io.kinesis: Amazon Kinesis data streams support.
• tensorflow_io.hadoop: Hadoop SequenceFile format support.
• tensorflow_io.arrow: Apache Arrow data format support. Usage guide here.
• tensorflow_io.image: WebP and TIFF image format support.
• tensorflow_io.libsvm: LIBSVM file format support.
• tensorflow_io.video: Video file support with FFmpeg.
• tensorflow_io.parquet: Apache Parquet data format support.
• tensorflow_io.lmdb: LMDB file format support.
• tensorflow_io.mnist: MNIST file format support.
• tensorflow_io.pubsub: Google Cloud Pub/Sub support.
• tensorflow_io.oss: Alibaba Cloud Object Storage Service (OSS) support. Usage guide here.

Installation

Python Package

The tensorflow-io Python package could be installed with pip directly:

$ pip install tensorflow-io

People who are a little more adventurous can also try our nightly binaries:

$ pip install tensorflow-io-nightly

The use of tensorflow-io is straightforward with keras. Below is the example of Get Started with TensorFlow with data processing replaced by tensorflow-io:

import tensorflow as tf
import tensorflow_io.mnist as mnist_io

# Read MNIST into tf.data.Dataset
d_train = mnist_io.MNISTDataset(
    'train-images-idx3-ubyte.gz',
    'train-labels-idx1-ubyte.gz')

# By default image data is uint8 so conver to float32.
d_train = d_train.map(lambda x, y: (tf.image.convert_image_dtype(x, tf.float32), y)).batch(1)

model = tf.keras.models.Sequential([
  tf.keras.layers.Flatten(input_shape=(28, 28)),
  tf.keras.layers.Dense(512, activation=tf.nn.relu),
  tf.keras.layers.Dropout(0.2),
  tf.keras.layers.Dense(10, activation=tf.nn.softmax)
])
model.compile(optimizer='adam',
              loss='sparse_categorical_crossentropy',
              metrics=['accuracy'])

model.fit(d_train, epochs=5, steps_per_epoch=10000)

Note that in the above example, MNIST database files are assumed to have been downloaded and saved to the local directory. Compression files (e.g. gzip) could be detected and uncompressed automatically.

R Package

Once the tensorflow-io Python package has beem successfully installed, you can then install the latest stable release of the R package via:

install.packages('tfio')

You can also install the development version from Github via:

if (!require("devtools")) install.packages("devtools")
devtools::install_github("tensorflow/io", subdir = "R-package")

TensorFlow Version Compatibility

To ensure compatibility with TensorFlow, it is recommended to install a matching version of TensorFlow I/O according to the table below:

TensorFlow I/O Version	TensorFlow Compatibility	Release Date
0.5.0	1.13.x	Apr 12, 2019
0.4.0	1.13.x	Mar 01, 2019
0.3.0	1.12.0	Feb 15, 2019
0.2.0	1.12.0	Jan 29, 2019
0.1.0	1.12.0	Dec 16, 2018

Developing

Python

For Python development, a reference Dockerfile here can be used to build the TensorFlow I/O package (tensorflow-io) from source:

$ # Build and run the Docker image
$ docker build -f dev/Dockerfile -t tfio-dev .
$ docker run -it --rm --net=host -v ${PWD}:/v -w /v tfio-dev
$ # In Docker, configure will install TensorFlow or use existing install
$ ./configure.sh
$ # Build TensorFlow I/O C++
$ bazel build -s --verbose_failures //tensorflow_io/...
$ # Run tests with PyTest, note: some tests require launching additional containers to run (see below)
$ pytest tests/
$ # Build the TensorFlow I/O package
$ python setup.py bdist_wheel

A package file dist/tensorflow_io-*.whl will be generated after a build is successful.

NOTE: When working in the Python development container, an environment variable TFIO_DATAPATH is automatically set to point tensorflow-io to the shared C++ libraries built by Bazel to run pytest and build the bdist_wheel. Python setup.py can also accept --data [path] as an argument, for example python setup.py --data bazel-bin bdist_wheel.

Starting Test Containers

Some tests require launching a test container before running. In order to run all tests, execute the following commands:

$ bash -x -e tests/test_ignite/start_ignite.sh
$ bash -x -e tests/test_kafka/kafka_test.sh start kafka
$ bash -x -e tests/test_kinesis/kinesis_test.sh start kinesis

Running Python Style Checks

Style checks for Python can be run with the following commands:

$ curl -o .pylint -sSL https://raw.githubusercontent.com/tensorflow/tensorflow/master/tensorflow/tools/ci_build/pylintrc
$ find . -name '*.py' | xargs pylint --rcfile=.pylint

R

We provide a reference Dockerfile here for you so that you can use the R package directly for testing. You can build it via:

docker build -t tfio-r-dev -f R-package/scripts/Dockerfile .

Inside the container, you can start your R session, instantiate a SequenceFileDataset from an example Hadoop SequenceFile string.seq, and then use any transformation functions provided by tfdatasets package on the dataset like the following:

library(tfio)
dataset <- sequence_file_dataset("R-package/tests/testthat/testdata/string.seq") %>%
    dataset_repeat(2)

sess <- tf$Session()
iterator <- make_iterator_one_shot(dataset)
next_batch <- iterator_get_next(iterator)

until_out_of_range({
  batch <- sess$run(next_batch)
  print(batch)
})

Community

• SIG IO Google Group and mailing list: io@tensorflow.org
• SIG IO Monthly Meeting Notes
• Gitter room: tensorflow/sig-io

License

Apache License 2.0