9 Things You Should Know About TensorFlow

Pinterest LinkedIn Tumblr

This morning I caught myself thinking that I was running through my head the fragments I liked from the What’s New with TensorFlow? at Google Cloud Next in San Francisco.

Then I thought about it for a moment and didn’t say to myself, why not share my ultra-short summary with you (unless you end up wanting to watch the video, and you need to do this: the lecturer is amazing), so we start .. …

# 1 It’s a powerful machine learning system

TensorFlow is a machine learning system that can be your new best friend if you have a lot of data and desire to comprehend the latest advancement in artificial intelligence: deep learning . Neural networks. Big ones. It’s not a data science magic wand, it’s a whole book of spells … which means you probably need to stop reading if you just want to plot a 20 by 2 table regression line.

But tremble if you want more. TensorFlow is being used to hunt new planets , prevent blindness, help doctors scan for diabetic retinopathy, and save forests by alerting authorities to signs of illegal logging . This is what AlphaGo and Google Cloud Vision are built on , and this is your theme. TensorFlow has source code, you can download it for free and get started right away .

TensorFlow’s discovery of Kepler-90i makes Kepler-90 the only known system with eight planets around a single star. We haven’t found another system with eight planets yet, so I’m guessing that means we’re still tied to Kepler-90. More details here .

# 2 The fancy approach is optional

I’m not myself from the TensorFlow Eager .

If you’ve tried TensorFlow in the old days and screamed out of it because you had to code as a scientist or as an alien and not as a developer, come back!

TensorFlow’s eager execution allows you to interact with it like a real Python programmer: you are given all the immediacy of writing and debugging code line by line, rather than stressful plotting. I myself used to be a scientist (and quite possibly an alien), but I was in awe of the instant performance of TF when it came out. Pleases instantly!

# 3 You can build neural networks one by one

Keras + TensorFlow = simplified neural network design!

Keras is the very convenience and simplicity of prototyping, TensorFlow has long needed them. If you like object oriented thinking and like building neural networks one layer at a time, you will love tf.keras. In a few lines of code below, we have created a sequential neural network with standard whistles and puffs in the form of a drop-down list (somehow I will fall into a lyrical mood and give you my metaphor about drop-down lists, it will have staplers and flu).

Oh, you like puzzles, don’t you? Patience. Don’t think too much about staplers.

# 4 It’s not just Python

Okay, you’ve been complaining about TensorFlow’s obsession with Python for a while now. Good news! TensorFlow isn’t just for pythons now. It now works in many languages, from R to Swift to JavaScript.

# 5 You can do everything in the browser

Speaking of JavaScript, you can train and execute models in the browser using TensorFlow.js. Chat with cool demos, I’ll wait here for your return.

Real-time human pose estimation in the browser using TensorFlow.js. Turn on your camera for a demonstration here. Or stay out of your chair. ¯ \ _ (ツ) _ / ¯ It’s up to you.

# 6 There is a Lite version for small devices

Have you taken your old computer to the museum? Got a toaster? (Anything as useless?) TensorFlow Lite allows the model to run on a variety of devices, including mobiles and the Internet of Things, giving you over 3x faster output than the original TensorFlow. Yes, you can now get machine learning on your Raspberry Pi or on your phone. In his presentation, Lawrence boldly shows the classification of images on an Android emulator in front of thousands of people … and it works.

1.6 seconds to compute? Yes! Banana more than 97% likely? Yes! Toilet paper? Well, I’ve been to several countries where a piece of paper like the one Lawrence is holding is considered toilet paper.

# 7 Specialized equipment has improved

If you are tired of waiting for your processor to finish working with your data to train your neural network, you can now use your hardware specially designed to work with cloud TPUs . T means tensor. Like TensorFlow … a coincidence? I don’t think so! A few weeks ago, Google announced the TPU versions in version 3.

# 8 New data streams have been greatly improved

What are you doing with numpy ? If you want to do it in TensorFlow, but get angry and give up, the tf.data namespace now makes your handling of input in TensorFlow more expressive and efficient. tf.data gives you fast, flexible and easy-to-use data pipelines synchronized with training.

# 9 You don’t have to start from scratch

You know it’s not very fun to get started with machine learning? Blank new page in your editor and no sample code for miles around. With TensorFlow Hub, you can use a better version of the time-honored tradition of borrowing someone else’s code and calling it your own (otherwise known as professional software development).

TensorFlow Hub is a repository for reusable pre-trainable machine learning model components packaged for one-time reuse. Help yourself!

Original Source