This repo contains the following:

• Source code for running automated machine learning using the python TPOT framework.
• A Dockerfile for building a Docker container to run a python3.4.6 environment.

Once volume mounts have been properly configured, it should be possible to launch the container and TPOT process simply by executing the run.sh script.

This github repository has been linked to a docker hub repository allowing the container to be automatically built following each push on master to github. The latest container will be pulled automatically from docker hub when the run.sh script is run. More specifically when a docker run or docker pull command is issued.

During experimentation, you may also want to build the container locally (or remotely), this can be done with:

# Where <tag> is the version number or nominally 'latest'

docker build -t ukgovdatascience/govuk-taxonomy-ml-image:<tag> .

No data are stored in the docker image itself, so these must be made available to the container by mounting a volume. This is done automatically in the run.sh, which connects the $DATADIR location with an internal mount point on the docker container (/mnt/data). $DATADIR should thus point at the location on the local machine where the data are stored.