search

tensorflow / fold

Deep learning with dynamic computation graphs in TensorFlow
Apache License 2.0
1.82k stars 266 forks source link

Compatability with Tensorflow 2.4+ #105

Open muraliadithya opened 3 years ago

muraliadithya commented 3 years ago

I am interested in using Tensorflow Fold, but I would like for my application to be extensible and reproducible. Are there plans to upgrade this repo to more recent versions of TF? If not, how extensive is the task of doing the upgrading myself? I am willing to work on a PR for that.

delesley commented 3 years ago

As you have noticed, the Tensorflow Fold repository is not being actively updated right now, in part because there are no users. :-) In the past, when I pushed an update, it was relatively trivial to do; I simply changed the hash code for the tensorflow directory to the latest version and recompiled. The Loom library in particular (which is what I'd recommend using) is relatively self-contained, and should be easy to port. The switch to python 3 required a few tricky changes that I (unfortunately) haven't pushed to github, and I haven't attempted to update the API to tensorflow 2.0, which will require more than just a recompile.

Can you give me a brief description of what you want to use TensorFlow Fold for?

-DeLesley

On Fri, Mar 5, 2021 at 4:48 PM Adithya Murali @.***> wrote:

I am interested in using Tensorflow Fold, but I would like for my application to be extensible and reproducible. Are there plans to upgrade this repo to more recent versions of TF? If not, how extensive is the task of doing the upgrading myself? I am willing to work on a PR for that.

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/tensorflow/fold/issues/105, or unsubscribe https://github.com/notifications/unsubscribe-auth/ABQZWTLFKYSNXYLBFWMP6KLTCF3XTANCNFSM4YWEKCKQ .

-- DeLesley Hutchins | Software Engineer | @.*** | 505-206-0315

digital-idiot commented 3 years ago

Hi @delesley, I was looking for a Tensorflow based framework to perform some experiments with Recursive Neural Network. I am not exactly working with word embeddings as such but I am interested in the variable input shape part. For example I was wondering if it would be possible to create a layer which accepts input with dynamic length but generates features with predefined length. I am not sure whether fold can help me to do what I intend to but I want to start somewhere. Thanks for your contribution.

delesley commented 3 years ago

TensorFlow Fold can certainly help with recursive neural networks. You simply write a recursive python function, and it will handle the autobatching. The library that you want to use is the Loom library, not the Blocks library. There's a calculator example there that shows how to do a recursive traversal of an expression tree.

On Fri, Mar 12, 2021 at 4:58 AM Abhisek Maiti @.***> wrote:

Hi @delesley https://github.com/delesley, I was looking for a Tensorflow based framework to perform some experiments with Recursive Neural Network. I am not exactly working with word embeddings as such but I am interest in the variable input shape part. For example I was wondering if it would be possible to create a layer which accepts input with dynamic length but generates features with predefined length. I am not sure whether fold can help me to do what I intend to but I want to start somewhere. Thanks for your contribution.

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/tensorflow/fold/issues/105#issuecomment-797473977, or unsubscribe https://github.com/notifications/unsubscribe-auth/ABQZWTIK6AGKAWC4MXLYCCLTDIFYFANCNFSM4YWEKCKQ .

-- DeLesley Hutchins | Software Engineer | @.*** | 505-206-0315