ElasticDL: Developer’s Guide

Get the Source Code

You can get the latest source code from Github:

git clone https://github.com/sql-machine-learning/elasticdl
cd elasticdl

Development Tools in a Docker Image

We prefer to install all building tools in a Docker image.

scripts/travis/build_images.sh

Check Code Style

The above Docker image contains pre-commit and hooks. We can run it as a container and mount the local Git repo into the container. In this container, we run the pre-commit command.

docker run --rm -it -v $PWD:/work -w /work elasticdl:dev \
  bash -c "make -f elasticdl/Makefile && pre-commit run -a"

If you have all required tools installed, you can also run the same script on your host.

make -f elasticdl/Makefile && pre-commit run -a

Build Wheel Packages Using Your Modified Code

After modifying code, you can build wheel packages by running the following command in the root of the project.

scripts/docker_build_wheel.sh

If you have elasticdl:dev image, you can only run the following command to generate wheel packages in the root of the project.

docker run --rm -it --net=host -v "$PWD":/work -w /work elasticdl:dev \
    bash -c "scripts/build.sh"

You can find wheel packages in the build directory.

Build a Docker Image using Local Wheel Packages

After building wheel packages, you can build a docker image by elasticdl zoo.

elasticdl zoo init \
  --base_image=elasticdl:dev \
  --model_zoo=model_zoo \
  --local_pkg_dir=./build \

elasticdl zoo build --image=elasticdl:dev_test .

Then, you can submit a job using the image like the tutorial.