Reference for ultralytics/utils/callbacks/mlflow.py#

def sanitize_dict(x: dict) -> dict:
    """Sanitize dictionary keys by removing parentheses and converting values to floats."""
    return {k.replace("(", "").replace(")", ""): float(v) for k, v in x.items()}

def on_pretrain_routine_end(trainer):
    """Log training parameters to MLflow at the end of the pretraining routine.

    This function sets up MLflow logging based on environment variables and trainer arguments. It sets the experiment
    name and run name, then starts the MLflow run if not already active. It finally logs the parameters from the
    trainer.

    Args:
        trainer (ultralytics.engine.trainer.BaseTrainer): The training object with arguments and parameters to log.

    Notes:
        MLFLOW_TRACKING_URI: The URI for MLflow tracking. If not set, MLflow uses its own default local store.
        MLFLOW_EXPERIMENT_NAME: The name of the MLflow experiment. If not set, defaults to trainer.args.project, or
            '/Shared/Ultralytics' if no project is set.
        MLFLOW_RUN: The name of the MLflow run. If not set, defaults to trainer.args.name.
        MLFLOW_KEEP_RUN_ACTIVE: Whether to keep the MLflow run active after training ends. Truthy values are
            "1", "true", "yes", "on", "y", "t" (case-insensitive); anything else is False.
    """
    if TESTS_RUNNING and "test_mlflow" not in os.environ.get("PYTEST_CURRENT_TEST", ""):
        return  # do not log during unrelated pytest tests

    # Set experiment and run names
    experiment_name = os.environ.get("MLFLOW_EXPERIMENT_NAME") or trainer.args.project or "/Shared/Ultralytics"
    run_name = os.environ.get("MLFLOW_RUN") or trainer.args.name

    trainer._mlflow_active = False
    trainer._mlflow_started_run = False
    try:
        mlflow.set_experiment(experiment_name)
        mlflow.autolog()
        active_run = mlflow.active_run()
        if active_run is None:
            active_run = mlflow.start_run(run_name=run_name)
            trainer._mlflow_started_run = True
        LOGGER.info(f"{PREFIX}logging run_id({active_run.info.run_id}) to {mlflow.get_tracking_uri()}")
        LOGGER.info(f"{PREFIX}disable with 'yolo settings mlflow=False'")
        mlflow.log_params(dict(trainer.args))
        trainer._mlflow_active = True
    except Exception as e:
        LOGGER.warning(f"{PREFIX}Failed to initialize: {e}")
        LOGGER.warning(f"{PREFIX}Not tracking this run")
        on_train_end(trainer, status="FAILED")

def _log_metrics(trainer, metrics):
    """Log metrics to MLflow, disabling tracking for this run on failure so it never crashes training."""
    try:
        mlflow.log_metrics(metrics=metrics, step=trainer.epoch)
    except Exception as e:
        LOGGER.warning(f"{PREFIX}metric logging failed, disabling tracking for this run: {e}")
        trainer._mlflow_active = False

def on_train_epoch_end(trainer):
    """Log learning rates and training losses at the end of each train epoch to MLflow."""
    if mlflow and getattr(trainer, "_mlflow_active", False):
        _log_metrics(
            trainer,
            {
                **sanitize_dict(trainer.lr),
                **sanitize_dict(trainer.label_loss_items(trainer.tloss, prefix="train")),
            },
        )

def on_fit_epoch_end(trainer):
    """Log validation metrics at the end of each fit epoch to MLflow."""
    if mlflow and getattr(trainer, "_mlflow_active", False):
        _log_metrics(trainer, sanitize_dict(trainer.metrics))

def on_train_end(trainer, status="FINISHED"):
    """Log artifacts for finished training and close any run this callback opened with its outcome status."""
    if not mlflow:
        return
    if status == "FINISHED" and getattr(trainer, "_mlflow_active", False):
        try:
            mlflow.log_artifact(str(trainer.best.parent))  # log save_dir/weights directory with best.pt and last.pt
            for f in trainer.save_dir.glob("*"):  # log all other files in save_dir
                if f.suffix in {".png", ".jpg", ".csv", ".pt", ".yaml"}:
                    mlflow.log_artifact(str(f))
            LOGGER.info(
                f"{PREFIX}results logged to {mlflow.get_tracking_uri()}\n{PREFIX}disable with 'yolo settings mlflow=False'"
            )
        except Exception as e:
            LOGGER.warning(f"{PREFIX}failed to log artifacts: {e}")
    if getattr(trainer, "_mlflow_started_run", False):  # only close a run we created
        trainer._mlflow_started_run = False
        if env_bool("MLFLOW_KEEP_RUN_ACTIVE"):
            LOGGER.info(f"{PREFIX}mlflow run still alive, remember to close it using mlflow.end_run()")
        else:
            try:
                mlflow.end_run(status=status)
                LOGGER.debug(f"{PREFIX}mlflow run ended")
            except Exception:
                pass