Hands-on Tensorflow 2.0: writing a GAN from scratch

4 June 2019

About me

Computer engineer | Head of ML & CV @ ZURU Tech, Italy | Machine Learning GDE

Blog: pgaleone.eu
Github: github.com/galeone
Twitter: @paolo_galeone

A glimpse at Tensorflow 2.0

A glimpse at Tensorflow 2.0

Tensorflow 1.x has a steep learning curve, because it uses dataflow graphs to represent computation.

If you come from an imperative programming language (C, Python, ...) you're not used to think as Tensorflow 1.x needs:

1. Define the computational graph, first.
2. Execute the described computation inside a Session.

In Tensorflow 2.0 the eager mode will be the default.

Eager mode

TensorFlow eager execution is an imperative programming environment that evaluates operations immediately, without building graphs: operations return concrete values instead of constructing a computational graph to run later.

• In Tensorflow 1.x it is possibile to enable eager mode.
• In Tensorflow 2.x it is possible to convert eager-mode code in Graph code (using tf.function).

In short: it is now possibile to use Tensorflow as a replacement for the most common numpy operations.

It's just eager mode enabled by default?

Of course not.

There will be a lot of changes, most of them will require a complete rewriting of the existing codebases.

• Transition from a graph-based way of thinking to an Object Oriented approach: from tf.layers to Keras

API clean-up

• Complete removal of tf.contrib and creation of separate projects
• Removal of deprecated and duplicated APIs

Better software design

• Removal of Global Collectios, use objects.
• No more Sessions, use functions.

Building a GAN from scratch, the Tensorflow 2.0 way

Building a GAN from scratch, the Tensorflow 2.0 way

Theory

• Google Colab (https://bit.ly/2X2WE1L)

Practice

• Google Colab (https://bit.ly/2Y7hYmM)

Conclusion

Tensorflow 2.0 is still in the early stages of the development, and the API can still change. However:

• using the eager mode for fast prototyping, and
• tf.function to convert the computationally intensive
• the Object Oriented approach that Keras forces to use

it is possibile to develop Machine Learning applications really easily.

From the software engineering point of view, Tensorflow 2.0 is a huge imporvement.

Annuocement

I'm authoring a book about Tensorflow 2.0 and Neural Networks!

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

Hands-On Neural Networks with Tensorflow 2.0

Understanding the Tensorflow architecture, from static graph to eager execution, designing Deep Neural Networks.

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

If you want to receive an email when the book is out, subscribe to the newsletter!

pgaleone.eu/subscribe/

Thank you