TensorFlow 2.0 Changes

A brief overview of what's coming in TensorFlow 2.0.
Eager execution by default, simple tools for graph mode, more object-oriented design (e.g., Keras), big cleanup, and more!

* Translations:

[laliborio]

Hi, dear Aurélien! Passing by just to say that your ml book is just amazing. Please keep writing these art works!

[PerisMartin]

Thanks for the update! Looking forward to the new book!!

[PritishMishra]

I am currently 16 and I know python very-very well as soon as I started ML. I get to know about your Book and just bought the Book.... and I searched your name on YouTube and just landed on your channel. You don't know how much you have helped beginners like me !! It will be Really really Appreciated if you make more such videos !! Thanks :)

[vaporizer]

Your book is great to understand the basics of machine learning and latest in R&D. For me this book is a very good reference book. If I want to get deeper into a subject your book provides enough guidance where to look and to start. I am looking forward for your 2nd edition.

[stavrospapadopoulos]

Great Video. Thank you.
It would be interesting to also make a video with a more in depth comparison between pytorch and tensorflow 2.0.

[danieldaza]

Great video, as usual! I'll be looking forward for the new edition of your book.

[JBoya]

Thanks for the very useful summary of the changes, and their impact beyond the coding syntax level.

[alexandrithsharron]

Thank you! Great video! Can't wait to pick up the second edition!

[Akkefa]

Thank you for detail introduction to tensorflow 2.0 new features.

[inflationova]

PyTorch 1.0 also brings optimizations in term of JIT and hybrid front-end. It would be very interesting to see how they compete each other.

[sidim.aourid]

Thank you for the video. Looking forward for your new book.

[jaircg]

Can't wait to read the new version of your book

[mdrayedbinwahed]

Fantastic video. I think it will greatly help beginners decide which platform to use.

[callum]

The citation graph @ 1 min is potentially quite misleading. A lot of papers will refer to caffe / caffe2 separately - I wonder if this was caught in the citation search. It is also worth noting that PyTorch and caffe2 are now effectively the same framework. Caffe2 was merged into PyTorch and is now used as the back end.

[citiblocsMaster]

TF 1.0 => TF 2.0 vs PyTorch is the exact same story as Angular 1.0 => Angular 2.0 vs React. We all know how it ends

[Skythedragon]

Nice!
Thanks for the info
Now Ill have to learn tf2.0 too!

[youserega]

This is great analysis, although I didn't like the balance scale, especially with TF 2.0 not even released yet. I also agree with others that frequently changing API is off-putting.

[Deesox]

An other excellent video! Thanks for it

[ami]

what about the data feed? for me that was the most painful place in using tensorflow

[geoffreyanderson]

Forecast for year 2019: Obsolete TF "answers" in code that does not work in TF2.0 still presented on StackOverflow after any Google search. TF tutorials still don't show anyone how to save and load models effectively. Nobody on Earth -- at least who publishes anything rather than keeping competitive secrets from corporate rivals -- knows how to use the "new" TF according to Google. Regurgitations of essentially the TF2.0 tutorial code, which is woefully incomplete in regard to real world deep learning project management, is widely re-written without attribution to original authors and without any evidence of working using reproducible research, except to buff the ego of fakers to their potential employers, and to get more hits from the world's most sub-par search engines.