Release TensorFlow 1.4

TensorFlow version 1.4 is now publicly available-this is a big update. We are very pleased to announce some exciting new features here and hope you enjoy it.

Keras

In version 1.4, Keras has migrated from Tf.contrib.keras to the core package Tf.keras. Keras is a very popular machine learning framework that contains a number of advanced APIs that can minimize the time between your creativity and your achievable implementation.

Keras can be integrated smoothly with other core tensorflow functions, including the estimator API. In fact, you can call the Tf.keras.estimator.model_to_estimator function to build the estimator directly from any Keras model. Because Keras is now added to the TensorFlow core, you can rely on it in the production workflow.

Data sets

We are pleased to announce that the Dataset API has migrated from Tf.contrib.data to the core package tf.data. Version 1.4 of the Dataset API also adds support for the Python builder. We strongly recommend that you use the Dataset API to create input pipelines for the TensorFlow model because:

The Dataset API can provide more functionality than the old API (feed_dict or queue-style pipelines).

The Dataset API is more performance-efficient.

The Dataset API is more concise and easier to use.

In the future, we will focus our development focus on the Dataset API rather than on the old API.

Distributed training and evaluation of estimators

The 1.4 edition also introduces a utility function tf.estimator.train_and_evaluate, which simplifies the export of training, evaluation, and estimator models. This function enables distributed execution of training and evaluation while still supporting local execution.

Other enhancements

In addition to the features described here, version 1.4 also introduces a number of other enhancements, which are described in the release notes:

Https://github.com/tensorflow/tensorflow/blob/master/RELEASE.md

Install TensorFlow 1.4

TensorFlow version 1.4 can now be obtained using standard PIP installation.

# note:the following command would overwrite any existing TensorFlow
# installation.
$ pip Install--ignore-installed--upgrade tensorflow
# Use PIP to Python 2.7
# use PIP3 instead ' pip for Pytho N 3.x

We have updated the documentation on tensorflow.org to 1.4.

The enhancement of TensorFlow is inseparable from contributors. Thank you very much for everyone involved in tensorflow development. Still hesitate what. Quickly join the community and develop the source code on the GitHub or help answer questions on Stack Overflow to become a contributor.

We want everyone to like all the features in this release.

I wish you all enjoy the TensorFlow code.

To view the full text and the link, please stamp the end of "read the original text."

Introducing immediate Execution for TensorFlow

AMP Roadshow technology Sharing roadshow China Special registration

The latest innovations in TensorFlow serving

Google releases TensorFlow lattice, prior knowledge drives flexibility