mlflow-gocd

GoCD plugins to work with MLFlow as model repository.

The plugin works with a process where runs within an experiment are "promoted" for production use. A new build is triggered for each promoted run in an experiment and exposes the artifact_uri as an environment variable to the build.

The optional fetch model plugin utilizes the artifacts uri and the given artifact pattern / name to download the model(s) (from S3 as of now) to the desired destination during the build.

Tested on GoCD 17.2.0+ and MLFlow 0.7.0

Configuring the repository

Configuring experiments as packages

Build comment with trackback url

Fetching models/artifacts from mlflow

Optionally the fetch plugin can also be used in conjunction with the package plugin to fetch artifacts stored in mlflow (backed by S3.)