Google details what’s coming in TensorFlow 2.0
A couple of years ago, nobody could have imagined that people would be talking about machine learning like they talk about the weather. The ever so powerful and complex technology has grown tremendously for the past two or three years. It managed to penetrate the mainstream technology industry and make everyone interested in its capabilities. One of the most popular machine learning-powered consumer-oriented platforms is TensorFlow, which was released about 3 years ago. Google has announced the official release of TensorFlow 2.0—the biggest update in the history of the platform. Here are some of the highlights from the upcoming update.
Easy model building
Last month, the TensorFlow team announced Keras—”a high-level API for building and training deep learning models.” The main advantage of Keras compared to the previous methods is that it’s user-friendly and offers modular and composable nature, which is easy to extend. Keras is aiming to make using the platform a much easier experience for new users.
Direct path to production
TensorFlow 2.0 has improved compatibility to make it easier to take your work straight to the production. You can now execute your models via TensorFlow Serving, TensorFlow Lite, or TensorFlow.js. Thanks to the TensorFlow’s broad community, you can use languages like C, Go, C#, Rust, and others.
More powerful training
TensorFlow 2.0 also includes new APIs like Keras Functional API and Model Subclassing to help you train your models “without sacrificing speed or performance.” Extensions like Ragged Tensors, TensorFlow Probability, and Tensor2Tensor help researchers experiment in a reliable environment.
The first public preview of 2.0 will be available early this year. Though, you can already start developing using tf.keras to make your projects future-proof. This is a powerful platform with tons of great tools. A more detailed changelog can be found in the Medium blog post below.