Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Good news, Tensorflow chooses Keras! #5050

Closed
kli-nlpr opened this Issue Jan 16, 2017 · 21 comments

Comments

Projects
None yet
@kli-nlpr
Copy link
Contributor

kli-nlpr commented Jan 16, 2017

Keras is gaining official Google support, and is moving into contrib, then core TF. If you want a high-level object-oriented TF API to use for the long term, Keras is the way to go.

http://www.fast.ai/2017/01/03/keras/

@faroit

This comment has been minimized.

Copy link
Contributor

faroit commented Jan 16, 2017

that would probably mean that theano support is dropped in near future.

Maybe @fchollet can speak about these plans here?

@hubayirp

This comment has been minimized.

Copy link

hubayirp commented Jan 16, 2017

@patyork

This comment has been minimized.

Copy link
Contributor

patyork commented Jan 16, 2017

@faroit @fchollet works for Google - pretty sure this is his initiative. I highly doubt he will be deleting the Keras github library here, purging all the theano code, and just moving the TF related code to the TF library.

@faroit

This comment has been minimized.

Copy link
Contributor

faroit commented Jan 16, 2017

I highly doubt he will be deleting the Keras github library here, purging all the theano code, and just moving the TF related code to the TF library.

Lets talk again in 1y, but I highly doubt this decision would allow both low level libs to be first class citizens in keras. There will be TF-only features that will not get backported to theano. And at one point the theano compatible version will significantly diverge from the tf-version so that it become too hard to maintain both.

@patyork

This comment has been minimized.

Copy link
Contributor

patyork commented Jan 16, 2017

@faroit In that case, you should probably start writing a theano replacement for keras now, before that happens.

Talk to you in a year.

@farizrahman4u

This comment has been minimized.

Copy link
Member

farizrahman4u commented Jan 16, 2017

Our contributions though... :(

@fchollet

This comment has been minimized.

Copy link
Collaborator

fchollet commented Jan 16, 2017

pretty sure this is his initiative

In fact this is not my initiative, it was decided by the TF leads at Google. But I do believe this is absolutely the way to go.

To set the record clear: Theano support will continue for as long as Keras exists, because Keras is meant as an interface rather than as an end-to-end framework. In fact, we will even expand the multi-backend aspect of Keras in the future: for instance Microsoft is working on a CNTK backend, and I'm excited about what comes out of that initiative. Keras is a front end for deep learning --not a front end for TensorFlow.

Our contributions though... :(

I don't understand this comment. Can you clarify? Our contributions what?

@faroit

This comment has been minimized.

Copy link
Contributor

faroit commented Jan 16, 2017

To set the record clear: Theano support will continue for as long as Keras exists, because Keras is meant as an interface rather than as an end-to-end framework. In fact, we will even expand the multi-backend aspect of Keras in the future: for instance Microsoft is working on a CNTK backend, and I'm excited about what comes out of that initiative. Keras is a front end for deep learning --not a front end for TensorFlow

@fchollet that is strong message to the community and for the devs here. Thanks! When the keras-tf-integration is finished this should def be something I'd put in a blogpost ;-)

@farizrahman4u

This comment has been minimized.

Copy link
Member

farizrahman4u commented Jan 16, 2017

I don't understand this comment. Can you clarify? Our contributions what?

What exactly would happen to this repository? Will this be the Keras home or the tensorflow repo? If tensorflow repo is the new "home" of Keras, then contributors of this repo will loose the credits right? Technically they would have contributions to this repo.. but this repo itself would be obsolete..

@faroit

This comment has been minimized.

Copy link
Contributor

faroit commented Jan 16, 2017

If tensorflow repo is the new "home" of Keras, then contributors of this repo will loose the credits right? Technically they would have contributions to this repo.. but this repo itself would be obsolete..

Also interesting would be to know how exactly these integration would legally look like. Contributions for tensorflow require to sign googles CLA. So would all keras contributors need to sign this?

@fchollet

This comment has been minimized.

Copy link
Collaborator

fchollet commented Jan 16, 2017

@farizrahman4u : nothing happens to this repository. It keeps being the home of Keras, and contributor credits aren't lost. What happens is that a TF-only version of the Keras interface will be added to TensorFlow (this works has already started). The two versions will be coupled but relatively independent.

Also interesting would be to know how exactly these integration would legally look like. Contributions for tensorflow require to sign googles CLA. So would all keras contributors need to sign this?

You don't need to sign anything to contribute to Keras, that doesn't change. As for any Keras code getting copied over to TensorFlow, that is possible because Keras and TensorFlow have interchangeable OSS licenses.

@dusenberrymw

This comment has been minimized.

Copy link

dusenberrymw commented Jan 19, 2017

Congratulations to you and the overall project, @fchollet! Keras is fantastic! 👍

@tstandley

This comment has been minimized.

Copy link

tstandley commented Jan 21, 2017

Does this mean we will have to choose between the Keras in the TF repository and this Keras? Will the TF people attempt to keep the two versions compatible?

@fchollet

This comment has been minimized.

Copy link
Collaborator

fchollet commented Jan 21, 2017

@ypxie

This comment has been minimized.

Copy link

ypxie commented Feb 20, 2017

@fchollet

Just curious, Is it possible to mix keras_tensor with native tensorflow functions without using lamda layer in this release?

@pianoman4873

This comment has been minimized.

Copy link

pianoman4873 commented Mar 4, 2017

@fchollet when will it be possible to use the keras in TF's contrib ? I've seen your demo at Google's dev summit and was very impressed by the api.

@sun9700

This comment has been minimized.

Copy link

sun9700 commented Mar 31, 2017

@fchollet What's your suggestion to TF beginner, TF.learn or TF.contrib.keras?
The TF offical tutorial is still focus on TF.learn. Will this tutorial be replaced?

@ahundt

This comment has been minimized.

Copy link
Contributor

ahundt commented Mar 31, 2017

@sun9700

This comment has been minimized.

Copy link

sun9700 commented Mar 31, 2017

will tf.contrib.keras be the offical first choice TF API?

@ahundt

This comment has been minimized.

Copy link
Contributor

ahundt commented Mar 31, 2017

fchollet/keras will be a spec, all other locations will be implementations of that spec

@stale stale bot added the stale label Jun 30, 2017

@stale

This comment has been minimized.

Copy link

stale bot commented Jun 30, 2017

This issue has been automatically marked as stale because it has not had recent activity. It will be closed after 30 days if no further activity occurs, but feel free to re-open a closed issue if needed.

@stale stale bot closed this Jul 30, 2017

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
You can’t perform that action at this time.