Run MLflow in Local or Remote Mode
This page provides two example MLflow workflows using scikit-learn linear regression models in a JupyterHub Notebook environment—one runs locally with the MLflow UI, and the other logs experiments to a remote MLflow tracking server.
Scenario 1: Local MLflow Tracking and UI Launch
Set up your environment
Install the required Python packages, if not installed.
pip install mlflow scikit-learn pandasRun a complete MLflow workflow
- Generate synthetic data using sklearn.datasets.make_regression.
- Split the data into training and test sets.
- Train a Linear Regression model using sklearn.linear_model.LinearRegression.
- Log model parameters, metrics, and the model to the local MLflow tracking server.
- Print the run ID and evaluation metrics for quick reference.
- Launch the MLflow UI to explore logged runs:
mlflow uiAccess the UI at http://localhost:5000 to view experiment results, compare runs, and manage artifacts.
Key Code Snippet
x
import subprocessimport mlflowimport mlflow.sklearnfrom sklearn.linear_model import LinearRegressionfrom sklearn.metrics import mean_squared_error, r2_scorefrom sklearn.model_selection import train_test_splitimport numpy as npnp.random.seed(42)X = np.random.rand(100, 1) * 10y = 3.5 * X.squeeze() + np.random.randn(100) * 2X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2)with mlflow.start_run(): model = LinearRegression() model.fit(X_train, y_train) predictions = model.predict(X_test) mse = mean_squared_error(y_test, predictions) r2 = r2_score(y_test, predictions) mlflow.log_param("model_type", "LinearRegression") mlflow.log_metric("mse", mse) mlflow.log_metric("r2", r2) mlflow.sklearn.log_model(model, "model") print("Run ID:", mlflow.active_run().info.run_id) print(f"MSE: {mse:.2f}, R²: {r2:.2f}")# Launch MLflow UI servermlflow_ui_cmd = [ "mlflow", "ui", "--backend-store-uri", "./mlruns", "--host", "10.100.11.63", "--port", "5000"]process = subprocess.Popen(mlflow_ui_cmd, stdout=subprocess.PIPE, stderr=subprocess.PIPE)Notes
- Local tracking data stored in
./mlruns. - MLflow UI accessible at <MLflow_URL>.
Scenario 2: Remote MLflow Server with Custom Artifact Location
Set up your environment
Make sure MLflow and required dependencies are installed:
pip install mlflow scikit-learnRun a remote MLflow workflow
- Set the tracking URI to point to your remote MLflow server.
mlflow.set_tracking_uri("http://<remote-host>:5000")- Create or use an existing experiment and define the artifact location (e.g., S3, NFS):
mlflow.set_experiment("my-experiment")- Train a Linear Regression model using
sklearn.linear_model.LinearRegression. - Log parameters, metrics, and the model to the remote MLflow tracking server.
- Print run details such as run ID and metrics for quick access.
This setup allows centralized tracking, easier collaboration, and persistent storage of experiments.
Key Code Snippet
import mlflowimport mlflow.sklearnfrom sklearn.linear_model import LinearRegressionfrom sklearn.model_selection import train_test_splitfrom sklearn.metrics import mean_squared_error, r2_scoreimport numpy as npmlflow.set_tracking_uri("http://10.100.11.10:5000")experiment_name = "LinearRegressionLocalArtifact"artifact_location = "file:/tmp/mlflow/artifacts"if not mlflow.get_experiment_by_name(experiment_name): mlflow.create_experiment(name=experiment_name, artifact_location=artifact_location)mlflow.set_experiment(experiment_name)np.random.seed(42)X = np.random.rand(100, 1) * 10y = 3.5 * X.squeeze() + np.random.randn(100) * 2X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2)with mlflow.start_run(run_name="linear-regression-demo"): model = LinearRegression() model.fit(X_train, y_train) y_pred = model.predict(X_test) mse = mean_squared_error(y_test, y_pred) r2 = r2_score(y_test, y_pred) mlflow.log_param("model_type", "LinearRegression") mlflow.log_metric("mse", mse) mlflow.log_metric("r2", r2) mlflow.sklearn.log_model(model, artifact_path="model") print("Logged to MLflow server at http://10.100.11.10:5000") print("Artifacts stored in:", artifact_location) print("Run ID:", mlflow.active_run().info.run_id)Notes
- Artifacts stored at the configured remote location.
- Update the tracking URI and artifact location as needed.
Summary
These examples show how to:
- Run MLflow Tracking and UI locally
- Log experiments to a remote MLflow server with custom artifact storage
- Train, evaluate, and log models to ensure reproducibility
Use these templates to seamlessly integrate MLflow into your machine learning projects for effective experiment tracking and model management.
Was this page helpful?
On This Page
Run MLflow in Local or Remote ModeScenario 1: Local MLflow Tracking and UI LaunchSet up your environmentRun a complete MLflow workflowKey Code SnippetNotesScenario 2: Remote MLflow Server with Custom Artifact LocationSet up your environmentRun a remote MLflow workflowKey Code SnippetNotesSummary