domino-experiment-tracking

To track ML experiments end-to-end in Domino, you use an MLflow-based Experiment Manager to log parameters, metrics, and artifacts across projects. Set a unique experiment name, enable autologging, and start a run to begin tracking.

Automatic logging for PyTorch and TensorFlow models in Domino is fully supported by the MLflow-based Experiment Manager. You simply enable autolog for your chosen framework before starting your training run to capture metrics automatically.

Comparing ML runs and managing model versions across projects is handled through the Experiment Manager's run comparison and model registry features. The MLflow-based integration enables consistent visualization and tracking of model versions within Domino.

Domino's experiment tracking seamlessly integrates with existing Domino projects and jobs. This integration enables end-to-end ML experiment workflows, allowing you to log artifacts and metrics directly from your scheduled or interactive jobs.

The best way to manually log artifacts and metrics during model evaluation is using the MLflow tracking APIs inside the Domino Experiment Manager. This allows you to record specific parameters, metrics, and output artifacts for each run.

Using MLflow for experiment tracking organizes ML experiments inside Domino, enabling consistent logging, comparison, and registry of models across projects. It provides automatic logging for popular frameworks and structured run comparison tools that manual logging lacks.