days
-4
-3
hours
0
-2
minutes
-2
-5
seconds
-5
-2
search
JAX Developers Puzzle deadline approaching: Play for a chance to win a complimentary JAX London 2018 ticket!
What’s new in this ML favorite?

TensorFlow 1.5: Streamlined execution, lightweight options for ML

Jane Elizabeth
TensorFlow 1.5
© Shutterstock / kirill_makarov

TensorFlow, the internet’s most popular machine learning project, is back. What does this latest update bring to our favorite platform for ML? TensorFlow’s newest features include updates for Eager Execution, TensorFlow Lite, and more!

TensorFlow is one of the most popular and celebrated machine learning projects currently out there. This week, the team behind this wildly popular machine learning project announced an update with the release of TensorFlow 1.5.

Let’s take a look at what’s in store!

Eager Execution ASAP

TensorFlow is getting even simpler with the addition of Eager Execution. While this is still only a preview feature for now, Eager Execution simplifies coding in TF. Developers have been fairly vocal about their desire for an imperative, define-by-run programming style. Eager Execution allows developers to execute TensorFlow operations as soon as they are called from Python.

The TF team offered a nice example for how this would work and look like on something like a simple computation for matrix multiplication. Originally, in TensorFlow 1.4, it would look like this:

x = tf.placeholder(tf.float32, shape=[1, 1])
m = tf.matmul(x, x)

with tf.Session() as sess:
  print(sess.run(m, feed_dict={x: [[2.]]})) 

But with Eager Execution and TensorFlow 1.5, it would look like this:

x = [[2.]]
m = tf.matmul(x, x)

print(m) 

The difference is pretty stark. If you want to learn more about Eager Execution, you can find more information here and here.

SEE MORE: Open source speech recognition toolkit Kaldi now offers TensorFlow integration

TensorFlow Lite

This update also brings more fun for developers who want to use ML on mobile platforms, thanks to the built-in developer version of TensorFlow Lite. (We covered this lightweight version of TensorFlow last year when it first came out here.) TensorFlow Lite is specifically designed to be lightweight and fast, perfect for on-device machine learning.

Here’s what we said last time:

TensorFlow Lite is designed to be lightweight, with a small binary size and fast initialization. It also supports a variety of platforms, including Android and iOS. And, in order to make the mobile experience better, it’s optimized for mobile devices with improved loading times and hardware acceleration.

Since TensorFlow Lite supports the Android Neural Networks API, it will be able to take advantage of new mobile devices with custom built hardware for ML. However, TensorFlow Lite falls back on optimized CPU execution if accelerator hardware isn’t available. So, it doesn’t matter whether your mobile device is specifically enabled for ML: your models will run regardless.

This new update to TensoreFlow Lite even includes an app to help developers get started.

SEE MORE: Top 5 open source machine learning projects 

Updates galore

It wouldn’t be an new release version without updates. So, in no particular order, here they are:

  • Improved documentation, including a brand-new how-to guide for beginners.
  • Improved support for NVidia CUDA 9 and cuDNN 7 for GPU Acceleration on Windows or Linux.
  • Improved support for Accelerated Linear Algebra (XLA)
  • Add streaming_precision_recall_at_equal_thresholds, a method for computing streaming precision and recall with O(num_thresholds + size of predictions) time and space complexity.
  • Change RunConfig default behavior to not set a random seed, making random behavior independently random on distributed workers. We expect this to generally improve training performance. Models that do rely on determinism should set a random seed explicitly.

SEE MORE: What’s new in TensorFlow 1.4?

Get it now

Interested in trying it out for yourself? TensorFlow 1.5 is available on GitHub here or through a pip installation. Happy coding!

Author
Jane Elizabeth
Jane Elizabeth is an assistant editor for JAXenter.com.

Leave a Reply

Be the First to Comment!

avatar
400
  Subscribe  
Notify of