Domino Experiment Tracking Skill

This skill provides comprehensive knowledge for tracking ML experiments in Domino Data Lab using the built-in MLflow-based Experiment Manager.

Key Concepts

Experiment Manager Overview

Domino's Experiment Manager is built on MLflow and provides:

• Automatic and manual logging of parameters, metrics, and artifacts
• Run comparison and visualization
• Model versioning and registry
• Integration with Domino projects and jobs

Critical Configuration

Experiment names must be unique across the entire Domino deployment. Always append username or project name to ensure uniqueness.

Related Documentation

• MLFLOW-BASICS.md - Auto-logging, manual logging
• COMPARING-RUNS.md - Run comparison, export
• MODEL-REGISTRY.md - Model registration & stages

Quick Start

import mlflow
import os

# CRITICAL: Experiment names must be unique across Domino deployment
username = os.environ.get('DOMINO_STARTING_USERNAME', 'unknown')
experiment_name = f"my-experiment-{username}"

# Set the experiment
mlflow.set_experiment(experiment_name)

# Enable auto-logging (easiest approach)
mlflow.autolog()

# Run training
with mlflow.start_run(run_name="my-first-run"):
    model.fit(X_train, y_train)

    # Optional: manually log additional items
    mlflow.log_param("custom_param", "value")
    mlflow.log_metric("custom_metric", 0.95)

Supported Frameworks

Environment Variables

Domino automatically configures MLflow to use the built-in tracking server. These variables are pre-set:

Documentation Links

• Domino Experiment Tracking: https://docs.dominodatalab.com/en/latest/user_guide/da707d/track-and-monitor-experiments/
• Domino Model Registry: https://docs.dominodatalab.com/en/latest/user_guide/3b6ae5/manage-models-with-model-registry/