Using the BucketFS Artifact Store

See the MLflow documentation for specifying the BucketFS Artifact Store either when starting the MLflow server or when creating an MLflow experiment.

As Default Artifact Repository

The following command line starts an MLflow server with the BucketFS as the default artifact store:

EXA_BUCKETFS_PASSWORD="<your password>" \
mlflow server --default-artifact-root \
exa+bfs://localhost:2580/bfsdefault/default/

This option is only available if you have access to the MLflow server and can change its startup options.

For more details, see URI Format of Artifact URIs.

In the Scope of an Individual MLflow Experiment

If you cannot change the startup options of your MLflow server, then you still can use the BucketFS Artifact Store for individual MLflow experiments.

Creating an MLflow Experiment describes how to create an MLflow experiment via UI, CLI, and API.

As soon as such an experiment exists, you can use it via MLflow API function set_experiment() providing the name or the ID of the experiment as argument.

import mlflow
import sklearn

mlflow.set_experiment("My Experiment")
model = sklearn.linear_model.LogisticRegression()
info = mlflow.sklearn.log_model(model, name="My_Model")
print(f"stored model at {info.artifact_path}")

For details, see the MLflow API function set_experiment().