Mlflow supportΒΆ
If you use MLflow and kedro-mlflow for the Kedro pipeline runs monitoring, the plugin will automatically enable support for:
starting the experiment when the pipeline starts,
logging all the parameters, tags, metrics and artifacts under unified MLFlow run.
To make sure that the plugin discovery mechanism works, add kedro-mlflow
as a dependencies to src/requirements.in
and run:
$ pip-compile src/requirements.in > src/requirements.txt
$ kedro install
$ kedro mlflow init
Then, adjust the kedro-mlflow configuration and point to the mlflow server by editing conf/local/mlflow.yml
and adjusting mlflow_tracking_uri
key. Then, build the image:
$ kedro docker build
And re-push the image to the remote registry.
If
kedro-mlflow
is not installed as dependency and configuration is not in place (missingkedro mlflow init
), the MLflow experiment will not be initialized and available for pipeline tasks in Apache Airflow DAG.