Debugging rTorch with Docker and Travis

27 August 2020

These are some instructions to build and run a Travis machine locally. We attempt to run all the tests usually run at in a Docker container maintaining the same characteristics of a Travis remote machine.


The goal is reducing the time that takes debugging an error in Anaconda environments when installing PyTorch or rTorch. The bugs remain hidden in the local development machine and only pop up in a Travis machine. That prevent us to complete the tests as the Travis tests can never be completed. There is always doubts of what bug is causing the problem.


There are some basic steps to reproduce a Travis remote machine and turn it into a local Docker container. First is identifying from our current Travis machine what is the Dockerhub instance that we should serve as a base of the local Travis machine. Second, we start writing a Dockerfile where the first line is downloading the remote image. Third, we add the locale, language, environment variables, and other settings. Fourth, we install the Linux dependencies,

Build Docker image

docker build -t my-travis .

Run container with travis

docker run --rm --name my-travis-kont -dit my-travis /sbin/init
docker exec -it my-travis-kont bash -l

Run PyTorch from inside the container

Get the PyTorch version


« Building the book "Statistical Rethinking" by Solomon Kurtz in Docker | rTorch - a PyTorch wrapper in R »