TensorFlow 2.0.0 release candidate: 2.0.0-rc2 includes breaking changes
We are one step closer to TensorFlow 2.0.0. The new release candidate, 2.0.0-rc2 includes new features, improvements, breaking changes, and bug fixes. Catch up on what’s expected to arrive in TensorFlow 2.0. The next major release focuses on ease of use and simplicity, with plenty of updates and easy model building with Keras.
TensorFlow is one of the most commonly used machine learning platforms. According to StackOverflow’s 2019 Developer Survey, it is many more times popular than Torch/PyTorch, and ranked as one of the most loved developer tools.
Originally developed ba the Google Brain team, now its GitHub description self-proclaims itself as the “open source machine learning framework for everyone“.
Its upgrade to version 2.0 included a focus on simplicity and ease of use. Currently in beta, 2.0 is a major new release. (It offers a place to start for beginners and experts alike.)
Teaming up
Going forward, TensorFlow and the API spec Keras will work together hand to build and train models.
SEE ALSO: Top Python use cases: Bringing machine learning to the enterprise
Read about the latest Keras release, version 2.3.0. It marks the last major release of multi-backend Keras. In the future, users are recommended to switch their code over to tf.keras in TensorFlow 2.0. tf.keras is included with TensorFlow, so users will not have to download it separately.
Keras brings model-building APIs including Sequential, Functional, and Subclassing.
Find out more about standardizing on Keras here.
TensorFlow 2.0.0-rc2
Protecting AI Solutions From Attacks
Alexander Polyakov (Independent Researcher)
AI for Decision Makers
Uri Eliabayev (Machine & Deep Learning Israel)
The new release candidate for TensorFlow includes improvements and new features.
Here are just a few of the highlights:
tf.distribute.StrategyAPI: Distribute models training with only a few code changes. This can be used with tf.keras and tf.estimator. Easily switch between strategies, including MirroredStrategy, TPUStrategy, MultiWorkerMirroredStrategy, CentralStorageStrategy, and ParameterServerStrategy.- More consistent API. API symbols receive renaming and changes. View the full list of changes.
- API clean-up and removal of
tf.app,tf.flags, andtf.logging - AutoGraph: Write regular Python inside
tf.function-decorated functions - TensorFlow API reference autocomplete fixes
- SavedModel unification of all ecosystem projects
- No more global namepsaces
- Easy model building using Keras
- Graph collections removal
Breaking changes and bug fixes
As with any new release, this release candidate comes with bug fixes and some backward-incompatible API changes.
Since TensorFlow 2.0.0 is built using GCC7 on Ubuntu 16, this has a possibility of leading to ABI incompatibilities with extensions using earlier versions. Users who experience issues should join the community testing group in order to help the devs. (Read the introductory post before posting.)
Make sure you take a look at all the breaking changes, listed here on GitHub.
Tracking v2.0
Refer to the roadmap for the full future plan.
SEE ALSO: Machine learning – How do Chess Engines work?
What else is coming in TensorFlow 2.0? Read the full blog on Medium written by the TensorFlow team and catch up on what changes are happening.
In the meantime, keep an eye on the public project status tracker and see how development is coming along. Currently, a final release date is not set.
