Some tips related to TensorFlow

Google Development Technology Extension engineer Laurence Moroney a 42-minute speech at Google Cloud Next Conference on the theme of "what's New with tensorflow?". The author Cassie Kozyrkov The lecture and summarizes nine things about TensorFlow. Machine Heart of this article was compiled to introduce, I hope to help you.

I've summed up my favorite speech at Google Cloud Next Conference--what's New with TensorFlow? (Https://www.youtube.com/watch?v=MR7w8ARAS2Y)

And then I thought, I couldn't find a reason not to give you Amway my short summary of this speech. (Unless you don't watch the video, you should definitely see this speech, the Speaker Laurence Moroney is excellent.) ）

#1: TensorFlow is a powerful machine learning framework

TensorFlow is a machine learning framework that, if you have a lot of data, or you are in the pursuit of Ai's most advanced technology: deep learning, neural networks, it will make you more powerful. It's not the Swiss Army Knife of the data science, but the industrial lathe! If all you want to do is draw a regression line through the 20X2 spreadsheet, you can stop reading this article.

But if you're looking for a bigger goal, then get up. ~tensorflow is used to find new planets, help doctors check for diabetic retinopathy to prevent blindness, and report illegal logging to the authorities to save the forest. It's the foundation of AlphaGo and Google Cloud Vision, and it's yours. TensorFlow is open source and you can download it for free and start using it now.

• tensorflow:https://www.tensorflow.org/install/
• TensorFlow Initial Tutorial: https://www.datacamp.com/community/tutorials/tensorflow-tutorial

With the help of TensorFlow, Kepler's -90i planet made Kepler-90 galaxies the only one we know of another eight planets orbiting a star. There are currently no more than eight planets in the galaxy, so I think that means the solar system and Kepler-90 galaxy tie (currently).

# # #: A magical operation

TensorFlow Eager gives me peace of mind.

If you have tried tensorflow before, but because it makes you like an old scholar or an alien (rather than a developer) coding and crazy, now hurry back AH ah Ah!!

TensorFlow's eager execution lets you interact like a pure Python programmer: Instant Writing and instant, progressive debugging, rather than holding your breath while building those huge charts. I am also a "scholar" (probably an alien) who is returning to normal, but I have fallen in love with TF's eager execution since it appeared. Strong Amway!

#3: Build neural networks line by row

Keras + TensorFlow = easier neural network build!

Keras is committed to user-friendliness and simple prototyping, which was desired by TensorFlow. If you like object-oriented thinking and like to build a layer of neural networks at once, you'll love Tf.keras. In the following lines of code, we created a sequence neural network (sequential neural networks) with standard ancillary components, such as dropout.

#4: not just Python

You complain that TensorFlow has been available for Python for some time. Now TensorFlow is no longer a patent for Python users. Now it can run in many languages, R, Swift, Java Script, and so on.

#5 you can do anything in the browser

Speaking of JavaScript, you can use Tensorflow.js to train and execute models in a browser. Let's try some cool cases here: https://js.tensorflow.org/~

Use Tensorflow.js to perform real-time human posture estimation in the browser. Open your camera and try it? Https://storage.googleapis.com/tfjs-models/demos/posenet/camera.html.

#6: Lite version for small devices

Bought an old table from the museum? A toaster? TensorFlow Lite makes it possible to execute models on a variety of devices, such as mobile devices and IoT devices, with a reasoning speed of more than 3 times times the original TensorFlow. Now you can do machine learning on a Raspberry Pi or mobile phone. In this speech, Laurence did a brave thing, in front of thousands of people with an Android simulator real-time demo image classification ... And it worked.

1.6 Seconds to calculate the time? Yes! Banana recognition rate over 97%? Yes!

#7: More robust dedicated hardware

If you're tired of waiting for the CPU to complete data processing while you're training a neural network, you can now use the hardware specifically designed for the Cloud TPU, T is tensor. Just like TensorFlow ... Is it a coincidence? I don't think so! Not long ago, Google released the third edition of TPU in the alpha version.

#8: New data workflows are greatly improved

What are you doing with NumPy? If you want to do the same thing in TensorFlow, but Rage-quit, the Tf.data namespace can help make input processing more expressive and efficient in tensorflow. Tf.data provides you with fast, flexible, and easy-to-use data workflows that are synchronized with training.

#9: You don't have to start from scratch

Do you know what's the scariest thing about starting machine learning? There is a new blank page in the editor, and there is not a lot of sample code. With TensorFlow Hub, you can execute a long-standing tradition more efficiently by writing someone else's code and calling it your own code (also known as professional software engineering).

TensorFlow Hub is a reusable, pre-trained machine learning model component repo, which is packaged and used very thinly. You can try it yourself!

• TensorFlow Official YouTube Channel: Https://www.youtube.com/channel/UC0rqucBdTuFTjJiefW5t-IQ
• Blog: https://medium.com/tensorflow

Some tips related to TensorFlow