deeplearning4j/dl4j-devkit repository overview

⁠dl4j-devkit

• the "devkit", a reference build environment for nd4j & deeplearning4j

⁠Contents

• OpenJDK 8
• Maven 3.3
• gcc
• blas

⁠Using

The devkit provides an interactive shell to build nd4j and deeplearning4j. The shell is packaged as a docker image.

⁠Quick Start

The following command will:

1. Start the devkit container.
2. Share the current directory into the container (at /workspace).
3. Clone the dl4j project to the /workspace directory.
4. Build the project.

docker run -it -v `pwd`:/workspace -w /workspace deeplearning4j/dl4j-devkit bash
$ git clone https://github.com/deeplearning4j/deeplearning4j.git
$ cd deeplearning4j
$ ./setup.sh

⁠Extended Use

⁠Install prerequisites

Mac OSX:

1. Install boot2docker:

brew install boot2docker

1. Install docker-compose:

pip install docker-compose

⁠Author a docker-compose file

To simplify launching the shell, author a docker-compose file into your workspace directory. Your workspace directory should contain the nd4j, deeplearning4j, and other such repositories.

Write to ./docker-compose.yml the following snippet:

shell:
  image: deeplearning4j/dl4j-devkit:latest
  volumes:
    - .:/workspace
    - $HOME/.m2:/home/vagrant/.m2
  working_dir: /workspace  
  environment:
    - CONTAINER_NAME=shell
  command: bash

⁠Start the shell

$ docker-compose run --rm shell

A bash shell starts up into the workspace directory that is shared into the container at /workspace.

[vagrant@abd582a75122#shell workspace]$ pwd
/workspace
[vagrant@abd582a75122#shell workspace]$ ls
Canova  deeplearning4j  dl4j-spark-ml nd4j

Build projects at will:

[vagrant@abd582a75122#shell workspace]$ cd nd4j
[vagrant@abd582a75122#shell nd4j]$ mvn install -DskipTests
...
[INFO] BUILD SUCCESS
...
[vagrant@abd582a75122#shell nd4j]$

Exit the shell when finished:

[vagrant@abd582a75122#shell workspace]$ exit