Skip to content

Dashboard Configuration

The Dashboard Configuration module allows you to select which dashboards are activated during model training. By default, the TensorBoard dashboard is always enabled and cannot be disabled.

WarpRec integrates with multiple dashboards through Ray, enabling flexible monitoring and logging of experiments.

Available Dashboards

The following nested sections can be used to configure dashboards:

  • wandb: Configuration for Weights & Biases.
  • mlflow: Configuration for MLFlow.
  • codecarbon: Configuration for CodeCarbon.

Important

  • Default state for all optional dashboards is disabled.
  • TensorBoard is always active and can be accessed locally.
  • Ensure all required credentials and API keys are valid; otherwise, dashboard logging may fail.

Weights & Biases (wandb)

WarpRec supports Weights & Biases via Ray for experiment tracking.

  • enabled: Boolean flag to activate the dashboard. Defaults to False.
  • team: Name of the team. If None, the first team is used. Defaults to None.
  • project: Name of the W&B project.
  • group: Name of the experiment group.
  • api_key_file: Path to a file containing the API key.
  • api_key: API key string.
  • excludes: List of parameters to exclude from logging.
  • log_config: Whether to log the experiment configuration. Defaults to False.
  • upload_checkpoints: Whether to upload model checkpoints. Defaults to False.

MLFlow (mlflow)

WarpRec supports MLFlow via Ray for experiment tracking and artifact logging.

  • enabled: Boolean flag to activate the dashboard. Defaults to False.
  • tracking_uri: URI of the MLFlow tracking server.
  • registry_uri: URI of the MLFlow model registry.
  • experiment_name: Name of the MLFlow experiment.
  • tags: Tags to attach to the MLFlow run.
  • tracking_token: Token for authentication with the MLFlow server.
  • save_artifacts: Whether to save artifacts to the MLFlow server. Defaults to False.

CodeCarbon (codecarbon)

WarpRec supports CodeCarbon via Ray to monitor energy consumption and carbon footprint.

  • enabled: Boolean flag to activate the dashboard. Defaults to False.
  • save_to_api: Whether to send results to the CodeCarbon API. Defaults to False.
  • save_to_file: Whether to save results locally to a file. Defaults to False.
  • tracking_mode: Tracking mode for CodeCarbon. Options are machine or process.

When using the save_to_api feature, the main to correctly track the experiment is to create a .codecarbon.config file which contains the following information:

[codecarbon]
experiment_id = YOUR_EXPERIMENT_ID
api_key = YOUR_API_KEY

Important

You can find the API key and the experiment ID inside the codecarbon official dashboard.

Example Dashboard Configuration

The following example shows a dashboard configuration enabling MLFlow tracking:

dashboard:
    mlflow:
        enabled: true
        experiment_name: MyExperiment

Note

Other dashboards (wandb, codecarbon) can be similarly enabled by specifying their corresponding nested sections. When using remote dashboards (e.g., W&B) with artifact logging, ensure that Ray checkpoints are not deleted during training.