72104 (1) [Avatar] Offline
#1
Hello
I am quite concerned about the previous version of tensorflow used throughout the current book draft (at least, the version that was current at the time chapters 1-9 of the book were released). It seems that the newly released tensorflow version 1.0 introduces new/changed APIs and other features. Most likely this entails changes in the presented APIs and example code.

Since the book is not released yet and most of us (readers) will or have already switched to version 1.0 very soon, it would be great to ensure that the examples and the whole book do not base on the obsolete tensor flow release - better said pre-release. Stated otherwise, it would be very welcome if text and code would be in sync with tensorflow version 1.0.

Of course, it can be said that text and code can be left as they are now and the reader has the possiblity to automatically convert scripts from the previous tensorflow version 0.1 to the newer 1.0. But this process is awkward at the very least and it would be far better if we could be sure that the book adapts to version 1.0, which seems to be a major release that will hopefully last for some time (years).

Can you please take the point seriously into account and give back your comments and plans?

regards antonio aranda
216292 (1) [Avatar] Offline
#2
+1. It would be great if the book covered 1.0. This would make this material more future-proof.

This is the reference to the conversation script: https://github.com/tensorflow/tensorflow/tree/master/tensorflow/tools/compatibility
455199 (2) [Avatar] Offline
#3
I second this point. By the time the bok comes out, the 0.12 version of tensorflow (which is the one followed in the book) will be already obsolete.

Requesting the author to convert to Tensorflow 1.0 in the next MEAP update, or at least in the final version.
125970 (1) [Avatar] Offline
#4
Perhaps people participating in the MEAP could post here in the forum where they run into incompatibilities.

The TensorFlow github repo has a tool to aid with upgrading at https://github.com/tensorflow/tensorflow/tree/r1.0/tensorflow/tools/compatibility -- I'm still working through examples in the book myself, but I will try to keep notes of where I hit snags and whether the upgrade utility script helps dig me out.
Daniel_ (5) [Avatar] Offline
#5
Yep plz, I would like to know if the book is going to be updated to TF 1.0. Otherwise, it will be my last MEAP book that I buy in Manning.
317976 (2) [Avatar] Offline
#6
It's not possible to use this meap with newer versions of tensorflow at all. Not just deprecation warnings, code in this book (right from the first chapters) doesn't work with tensorflow 1.0 at all.

The tensorflow api has gone through a lot of changes between the version in this book and the current one. It is in fact a dangerous idea to learn tensorflow from this book.

I hope they put in the effort to bring the book up to speed. It's not just that we meap readers are unhappy, I think most people wouldn't want to buy an outdated book on tensorflow.

I understand that frameworks and apis tend to change and evolve over time. But here there has been too much change, and the tensorflow code listings in this book are just not useful any more.
317976 (2) [Avatar] Offline
#7
My Bad!

There has just been an update to the book and at first glance it seems many things have been updated to match newer apis. I'll go through the code again and post my experience here.

I am glad and excited to see the effort.
454707 (2) [Avatar] Offline
#8
Hi:
Please, make changes to code of Chapter 10 (RNN), and I believe in some other parts in regard to sharing variables.

tf.get_variable_scope().reuse_variables() does not work, and I can not yet make it work with TF1.0 changes. It is so frustrating (((
394773 (1) [Avatar] Offline
#9
I agree. The book has to be updated with Tensorflow version 1.0 examples before it's final version is published, otherwise, this book will become obsolete before it is published. This happened for example to Spark in Action writers, they updated the book to have examples comtaible with last version before it was published.
Nishant Shukla (44) [Avatar] Offline
#10
Hi everyone!

Code in book and repo are all updated to TensorFlow v1.0! Enjoy! https://github.com/BinRoot/TensorFlow-Book