support for Tensorflow 2 about tfgo HOT 4 CLOSED

Official libs have been updated: https://www.tensorflow.org/install/lang_c

from tfgo.

tfgo already works with TensorFlow 2.x if you compile the C library by yourself, it's just a matter of linking it.

Officially I can't support it, because I have to wait for the official C library compiled. But as soon as it's out I will support it.

In the meantime, you can compile the C library by yourself (or, if like me you use Archlinux you can just install the tensorflow package and link with the ready-to-use .so) and use tfgo with the tf 2.x runtime.

from tfgo.

@mxk @Arnold1 sorry for the extreme delay (but lockdown + moving to a new apartment have absorbed all my time).

I'll have a look at TensorFlow 2.x support (I'll start with TensorFlow 2.3.1).
In order to be able to completely support it, I also have to rewrite the test part of tfgo (I use a Python script to generate a tensorflow 1.x support - now, instead, I have to export a tensorflow 2.x function (or model, or stati graph) using tf.function.

It won't be immediate, but perhaps if I find the time, in some weeks a new release should be available.

P.S: I covered the tf.function export + the load in tfgo in my book (https://amzn.to/2TMImSw) if you're interested, you can have a look at the last paragraph.

from tfgo.

I merged the support for TensorFlow 2.x (in particular, 2.3.1).

I had to remove the estimator support because TensorFlow is almost dropping them and are impossible to maintain alone.

Moreover, I had to create a TensorFlow fork with some fix in the tests and with all the required .proto files compiled and placed in the correct position, in this way it's easy to go build and also go get the package without any compilation error.

I updated the README accordingly

from tfgo.