TensorFlow interface for GNU Octave

Content:

1. About
2. Install
3. Contribute

This is an experimental Octave Package for interfacing GNU Octave to the TensorFlow library (v2.15.0) through its C API. At the moment, there is only a single dynamically linked octave library (the tensorflow function), which is handling the interface between Octave and TensorFlow's C API. This package is at a very early stage of development and the calls to the intermediate OCT interface might (most likely) change in the near future. Once the OCT interface supports most of TensorFlow's C API, Octave classdefs will be added as wrappers to facilitate the use of the TensorFlow library through GNU Octave. This package has been inspired by @asteinh's work on tensorflow.m.

Documentation for the tensorflow Octave Package can be found here.

The tensorflow OCT interface links against the TensorFlow library (v2.15.0) for CPU and GPU on Linux systems only. Other versions have not been tested. Before installing the tensorflow Octave Package, you must downloaded and install the required library.

Issuing the following commands from the Linux terminal will download, extract into /urs/local/lib, and make the library visible to the linker.

wget https://storage.googleapis.com/tensorflow/libtensorflow/libtensorflow-cpu-linux-x86_64-2.15.0.tar.gz
sudo tar -C /usr/local -xzf libtensorflow-cpu-linux-x86_64-2.15.0.tar.gz
sudo ldconfig /usr/local/lib

From within the Octave command, you can install the latest tensorflow package as usual with

pkg install -forge tensorflow

or install the latest developments (main) with

pkg install "https://github.com/pr0m1th3as/tensorflow/archive/refs/heads/main.zip"

As said, this is an experimental version under development. Few things have been tested so far, most things are likely to have bugs. Feel free to contribute with a PR or open an issue or both!