Update tensorflow version, please

[nikki-dzhurov]

I installed/linked the package but cannot run models that were built with tensorflow version > 1.3

Possible Unhandled Promise Rejection (id: 0):
05-26 21:42:48.749 6305 7014 W ReactNativeJS: Error: NodeDef mentions attr 'dilations' not in Op<name=Conv2D; signature=input:T, filter:T -> output:T; attr=T:type,allowed=[DT_HALF, DT_FLOAT]; attr=strides:list(int); attr=use_cudnn_on_gpu:bool,default=true; attr=padding:string,allowed=["SAME", "VALID"]; attr=data_format:string,default="NHWC",allowed=["NHWC", "NCHW"]>; NodeDef: conv1/Conv2D = Conv2D[T=DT_FLOAT, _output_shapes=[[64,48,48,96]], data_format="NHWC", dilations=[1, 1, 1, 1], padding="SAME", strides=[1, 2, 2, 1], use_cudnn_on_gpu=true](random_shuffle_queue_DequeueMany:1, conv1/kernel/read). (Check whether your GraphDef-interpreting binary is up to date with your GraphDef-generating binary.).

[sparkison]

This is b/c TensorFlow-experimental is currently at version 1.1. Currently, the only way to get a newer version of TF is to build it yourself

For now, your best be would be to use python virtual env and switch to TF 1.1 pip install --upgrade "tensorflow==1.1.0" then re-train your model

[nikki-dzhurov]

Thanks for your response.
I see your point, but I use some of the latest Tensorflow APIs. Unfortunately I cannot go back to version 1.1 so easily. I tried that with version 1.3 and it was a mess.
(btw I use Android for testing and I saw that there are new java libraries. I thought it would be easy for update)

[coire1]

Hello, I have the same problem: a trained model with TF 1.8.0 doesn't work with the package.
Could you tell me which commit of 'tensorflow/models' work with TF 1.1.0 or how can I train a model with TF 1.1.0? Unfortunately repo tags starts from 1.4.0. Thank you.

[shokimble]

See PR #28

....I just realised, almost midnight here that I have only included Android. ooops. I'll check IOS tomorrow. Probably very similar small change.

[shokimble]

Ok IOS is a bigger problem.

As @sparkison pointed out the official cocoapod from Google is waaaaaaay out of date. Over a year old.

You have to build from scratch as also pointed out but doing this as part of a manual installation seems a bit out of the scope of this library. Another option is to release a new Cocoapod with the latest version on it. I'm going to do that as soon as I get around to creating the IOS version of the app I'm running on but Android is good enough for me for now.

Using a third party's Cocoapod does come with trust issues. I could release a version full of malware and as it comes as a tar bundle with compiled libs most people would be unaware. Not that I'm planning to do so but it's important to note. No worse than npm I suppose apart from it being extremely difficult to tell what's in a binary library.

What I propose to update my pull request with is a link to a cocoapod hosted on my git account and updated README to say exactly what I said above as a warning but offering it as an option with instructions.

[reneweb]

Yep, the problem here is iOS, as Google doesn't seem to update the Tensorflow iOS lib at all anymore and only updating Android would greatly diverge iOS vs Android and I would prefer to keep them as close together as possible. Also worth noting that eventually I would like to move to Tensorflow Lite as per #10 - unfortunately Tensorflow Lite is still pretty young and there are a few issues with integrating it with React Native