package tensorflow

README.md

The tensorflow-ocaml project provides some OCaml bindings for TensorFlow.

Installation

Use opam to install the tensorflow-ocaml package.

opam install tensorflow

Get the TensorFlow Library

The opam package starting from version 0.0.8 requires the version 1.0 of the TensorFlow library to be installed on your system under the name libtensorflow.so. Two possible ways to obtain it are:

• You can build this library from source by following these steps:

  1. Install the Bazel build system

  2. Clone the TensorFlow repo.

  `git clone --recurse-submodules -b r1.0 https://github.com/tensorflow/tensorflow`
  

  1. In the tensorflow directory run ./configure (you will be asked if you want to enable CUDA support).

  2. Run bazel build -c opt tensorflow:libtensorflow.so.

  3. The resulting library should then be in bazel-bin/tensorflow/libtensorflow.so.

• You can download prebuilt binaries from Google: The releases are available in URLs of the form: https://storage.googleapis.com/tensorflow/libtensorflow/libtensorflow-TYPE-OS-ARCH-VERSION.tar.gz. For example:

  • CPU-only, Linux, x86_64, 1.0.0. https://storage.googleapis.com/tensorflow/libtensorflow/libtensorflow-cpu-linux-x86_64-1.0.0.tar.gz

  • GPU-enabled, Linux, x86_64, 1.0.0. https://storage.googleapis.com/tensorflow/libtensorflow/libtensorflow-gpu-linux-x86_64-1.0.0.tar.gz

  • CPU-only, OS X, x86_64, 1.0.0. https://storage.googleapis.com/tensorflow/libtensorflow/libtensorflow-cpu-darwin-x86_64-1.0.0.tar.gz

  • GPU-enabled, OS X, x86_64, 1.0.0: https://storage.googleapis.com/tensorflow/libtensorflow/libtensorflow-gpu-darwin-x86_64-1.0.0.tar.gz

Build a Simple Example

Download a very simple example and compile it with the following command:

ocamlbuild forty_two.native -pkg tensorflow -tag thread

Then run it via ./forty_two.native. You should now be all set up, enjoy!

Frequent Problems

• When compiling the example with ocamlbuild, I get the following error:

/usr/bin/ld: cannot find -ltensorflow

You should adjust your LIBRARY_PATH environment variable to include the directory in which you have added libtensorflow.so (and use this exact name). E.g. run:

LIBRARY_PATH=/path/to/lib:$LIBRARY_PATH ocamlbuild forty_two.native -pkg tensorflow -tag thread

• When running forty_two.native, I get the following error:

./forty_two.native: error while loading shared libraries: libtensorflow.so: cannot open shared object file: No such file or directory

You should adjust your LD_LIBRARY_PATH environment variable in the same way LIBRARY_PATH was adjusted in the previous case. E.g. run:

LD_LIBRARY_PATH=/path/to/lib:$LD_LIBRARY_PATH ./forty_two.native

Note that on OS X, you should adjust your DYLD_LIBRARY_PATH environment variable

Examples

Tensorflow-ocaml includes two different APIs to write graphs.

Using the Graph API

The graph API is very close to the original TensorFlow API.

• Some MNIST based tutorials are available in the examples directory.

• examples/load/load.ml contains a simple example where the TensorFlow graph is loaded from a file (this graph has been generated by examples/load.py).

• examples/basics contains some curve fitting examples. You will need gnuplot to be installed via opam to run the gnuplot versions.

Using the FNN API

The FNN API is a layer based API to easily build neural-networks. A linear classifier could be defined and trained in a couple lines:

  let input, input_id = Fnn.input ~shape:(D1 image_dim) in
  let model =
    Fnn.dense label_count input
    |> Fnn.softmax
    |> Fnn.Model.create Float
  in
  Fnn.Model.fit model
    ~loss:(Fnn.Loss.cross_entropy `mean)
    ~optimizer:(Fnn.Optimizer.gradient_descent ~learning_rate:8.)
    ~epochs
    ~input_id
    ~xs:train_images
    ~ys:train_labels;

A complete VGG-19 model can be defined as follows:

let vgg19 () =
  let block iter ~block_idx ~out_channels x =
    List.init iter ~f:Fn.id
    |> List.fold ~init:x ~f:(fun acc idx ->
      Fnn.conv2d () acc
        ~name:(sprintf "conv%d_%d" block_idx (idx+1))
        ~w_init:(`normal 0.1) ~filter:(3, 3) ~strides:(1, 1) ~padding:`same ~out_channels
      |> Fnn.relu)
    |> Fnn.max_pool ~filter:(2, 2) ~strides:(2, 2) ~padding:`same
  in
  let input, input_id = Fnn.input ~shape:(D3 (img_size, img_size, 3)) in
  let model =
    Fnn.reshape input ~shape:(D3 (img_size, img_size, 3))
    |> block 2 ~block_idx:1 ~out_channels:64
    |> block 2 ~block_idx:2 ~out_channels:128
    |> block 4 ~block_idx:3 ~out_channels:256
    |> block 4 ~block_idx:4 ~out_channels:512
    |> block 4 ~block_idx:5 ~out_channels:512
    |> Fnn.flatten
    |> Fnn.dense ~name:"fc6" ~w_init:(`normal 0.1) 4096
    |> Fnn.relu
    |> Fnn.dense ~name:"fc7" ~w_init:(`normal 0.1) 4096
    |> Fnn.relu
    |> Fnn.dense ~name:"fc8" ~w_init:(`normal 0.1) 1000
    |> Fnn.softmax
    |> Fnn.Model.create Float
  in
  input_id, model

This model is used in the following example to classify any input image, in order to use it you will have to download some pre-trained weights.

There are also some MNIST based examples.

Character level RNN

A simplified version of char-rnn can also be found in the examples directory which contains additional details.

Neural Style Transfer

A stand-alone example of Neural Style transfer can be found in the examples directory.

Dependencies

• ocaml-ctypes is used for the C bindings.

• Base is only necessary when generating the graph from OCaml, the wrapper itself does not need it.

• The code in the piqi directory comes from the Piqi project. There is no need to install piqi though.

• Cmdliner is used for command line interfaces.

• Gnuplot-ocaml is an optional dependency used by a couple examples.