Quick start

Requirements

To install and use tensorflow.m, you will need the following:

• a recent version of MATLAB or Octave (check the remarks on compatibility)

• a C compiler (set up to build the MEX file)

• internet access to download the TensorFlow libraries (if not already downloaded)

tensorflow.m does not depend on any MATLAB/Octave toolboxes or other third-party libraries.

Installation

Clone the repository to your preferred location

git clone git@github.com:asteinh/tensorflow.m.git rootdir

where we will refer to the root directory as rootdir.

tensorflow.m comes with a setup function that sets up the path, builds the MEX interface and generates TensorFlow operations. Fire up MATLAB/Octave, change your directory to rootdir and run setup.m. Especially the automatic generation of operations might take a while, so stay tuned.

If the setup finishes without any errors, you’re good to go on and run a first example.

First steps

We will now quickly introduce some key functionality of tensorflow.m. For more details, have a look at the API documentation.

More often than not, your implementation will start with a plain graph, which is initialized by

graph = tensorflow.Graph();

This graph subsequently allows us to use TensorFlow’s operations, e.g.

c = graph.constant(rand(2,3)); % creates a constant 2-by-3 tensor

To evaluate an operation, we need to create a session

session = tensorflow.Session(graph);

and then run the outputs we want to obtain

c_val = session.run([], [], c).value();

Note the appended .value(), which fetches the numeric values of a tensor.

What’s next

You’re all set. Feel free to head over to the examples and have a go at more complex examples, or start working on your own project. For details on available operations, etc. take a look at the API documentation.