[MLflow] Experiments

Create a MLflow experiment

Define the name of the experiment, and artifact_location, tags as default values for MLflow runs in the future.

import mlflow

if __name__ == "__main__":
    # create a new mlflow experiment
    mlflow.create_experiment(
        name = "MyFirstMLflow",
        artifact_location = "MyFirstMLflow_artifact_loc",
        tags={"env":"dev", "version":"1.0.0"}
    )

Make functions in a seperate file(mlflow_utils.py) and call the functions to create, retrive, and delete an experiment

The function creates an experiment only when there is no existing experiment with the name requested and returns the experiment_id

""" In mlflow_utils.py"""

import mlflow
from typing import Any

def create_mlflow_experiment(experiment_name: str, artifact_location: str, tags: dict[str,Any]) -> str:
    """
    Create a new mlflow experiment with the given name and artifact location
    """
    try:
        experiment_id = mlflow.create_experiment(
            name=experiment_name, artifact_location=artifact_location, tags=tags
        )
    except:
        print(f"Experiment {experiment_name} already exists.")
        experiment_id = mlflow.get_experiment_by_name(experiment_name).experiment_id
        
    return experiment_id

Now, you can call the function and create a new MLflow experiment.

from mlflow_utils import create_mlflow_experiment

if __name__ == "__main__":
    experiment_id = create_mlflow_experiment(experiment_name="MySecondMLflow",
                                              artifact_location="MySecondMLflow_artifact_loc",
                                              tags={"env":"dev", "version":"1.0.0"})
    print(f"Experiement ID: {experiment_id}")

Experiement ID: 977725533440491376

The function retrieves an experiment with arguments; experiment_id or experiment_name

""" In mlflow_utils.py"""

import mlflow
from typing import Any

def get_mlflow_experiment(
        experiment_id: str = None, experiment_name: str = None
    ) -> mlflow.entities.Experiment:
    """
    Retrieve the mlflow experiment with the given id or name.

    Parameters:
    ----------
    experiment_id: str
        The id of the experiment to retrieve.
    experiment_name: str
        The name of the experiment to retrieve.

    Returns:
    -------
    experiment: mlflow.entities.Experiment
        The mlflow experiment with the given id or name.
    """
    if experiment_id is not None:
        experiment = mlflow.get_experiment(experiment_id)
    elif experiment_name is not None:
        experiment = mlflow.get_experiment_by_name(experiment_name)
    else:
        raise ValueError("Either experiment_id or experiment_name must be provided.")
    return experiment

Retrieve an MLflow experiment and print the info

from mlflow_utils import get_mlflow_experiment

if __name__ == "__main__":
    # retrieve an MLflow experiment with its id
    experiment = get_mlflow_experiment(experiment_id="977725533440491376")

    # retrieve an MLflow experiment with its name
    # experiment = get_mlflow_experiment(experiment_name="MySecondMLflow")

    print("name: {}".format(experiment.name), "\n",
          "Experiment_id: {}".format(experiment.experiment_id), "\n",
          "Artifact location: {}".format(experiment.artifact_location), "\n",
          "Tags: {}".format(experiment.tags), "\n",
          "Lifecycle_stage: {}".format(experiment.lifecycle_stage), "\n",
          "Creation_time: {}".format(experiment.creation_time))

name: MySecondMLflow 
 Experiment_id: 977725533440491376 
 Artifact location: file:///c:/Users/dof07/Desktop/mlflow/MySecondMLflow_artifact_loc 
 Tags: {'env': 'dev', 'version': '1.0.0'} 
 Lifecycle_stage: active 
 Creation_time: 1741046879213

The function deletes an experiment with arguments; experiment_id or experiment_name

def delete_mlflow_experiment(
    experiment_id: str = None, experiment_name: str = None
) -> None:
    """
    Delete the mlflow experiment with the given id or name.

    Parameters:
    ----------
    experiment_id: str
        The id of the experiment to delete.
    experiment_name: str
        The name of the experiment to delete.
    """
    if experiment_id is not None:
        mlflow.delete_experiment(experiment_id)
    elif experiment_name is not None:
        experiment = mlflow.get_experiment_by_name(experiment_name)
        experiment_id = experiment.experiment_id
        mlflow.delete_experiment(experiment_id)
    else:
        raise ValueError("Either experiment_id or experiment_name must be provided.")

Delete an MLflow experiment

import mlflow

mlflow.delete_experiment(experiment_id="977725533440491376")
# mlflow.delete_experiment(experiment_name="MySecondMLflow")

NOTE : Creating an experiment outputs info folders and files sush as models(folder), experiment_id(folder), meta.yaml and tags(env, version as predefined)