undefined symbol: _ZN10tensorflow8internal21CheckOpMessageBuilder9NewStringEv on Linux about compression

[trougnouf]

Fixed. I just had to remove

        "-std=c++11",
        "-D_GLIBCXX_USE_CXX11_ABI=0",

from tensorflow_compression/cc/BUILD, then compile with bazel run -c opt :build_pip_pkg

[trougnouf]

I made an AUR/Arch package https://aur.archlinux.org/packages/python-tensorflow-compression-git/ with the patch on https://aur.archlinux.org/cgit/aur.git/tree/arch_compression.patch?h=python-tensorflow-compression-git

[jonycgn]

Hi, thanks for this.

If by "learning TF 1" you mean learning to write your model in graph mode, this will still be necessary unless you use the new entropy model implementations (under tensorflow_compression/python/entropy_models).

The EntropyBottleneck and GaussianConditional classes etc., which our examples and the documentation are based on, aren't compatible with eager mode (neither in TF1 nor in TF2, where eager mode is the default).