Key takeaways
- TensorFlow 2.13 has been released, following the previous update in March 2023.
- LMDB kernels have been modified to return errors in preparation for their removal from TensorFlow.
- Improvements in TF Lite include support for 16-bit and 64-bit float types and new interpreter options.
- The tf.data.Dataset.zip now supports Python-style zipping, enhancing dataset manipulation capabilities.
- Keras has removed deprecated API wrappers in favour of newer recommendations.
Google has released the latest version of its popular open-source software library, TensorFlow 2.13. The launch comes four months after the introduction of TensorFlow 2.12. In 2022, the team had released three versions – 2.8, 2.9, 2.10 and 2.11. For 2023, we already have the second update with the earlier one being released in March.
Check out the GitHub repository to learn more about the update.
In the updated version, the LMDB kernels have been changed to return an error. This is in preparation for completely removing them from TensorFlow. The LMDB dependency that these kernels are bringing to TensorFlow has been dropped, thus making the build slightly faster and more secure.
Other major improvements have been made to TF lite; the team has added 16-bit and 64-bit float type support for built-in op cast. Also, the Python TF Lite Interpreter bindings now have an option experimental_disable_delegate_clustering to turn-off delegate clustering.
Furthermore, the tf.data.Dataset.zip now supports Python-style zipping, i.e. Dataset.zip(a, b, c). Now ‘tf.data.Dataset.shuffle’ supports ‘tf.data.UNKNOWN_CARDINALITY’ when doing a "full shuffle" using dataset = dataset.shuffle(dataset.cardinality()). But a "full shuffle" will load the full dataset into memory so that it can be shuffled, so users are recommended to only use this while working with small datasets or datasets of small objects.
Similar to the last update, some considerable additions have been made to Keras as well. The team has removed the Keras scikit-learn API wrappers (KerasClassifier and KerasRegressor), which was deprecated in August 2021. Instead, the team recommended using SciKeras. They also added a utility to run a timed thread every x seconds. The feature can be used to run a threaded function alongside model training or any other snippet of code.
Other features like F-score metrics, activation function, experimental KPI for metrics and so on have been added. Though no update regarding the security of the software has been released.
