lazypredict.integrations Package

lazypredict.integrations.mlflow

mlflow

MLflow integration helpers for LazyPredict.

RunTracker

RunTracker(mlflow_module: Any = None)

Best-effort MLflow tracking during one fit().

Tracking never changes the benchmark: the first MLflow error is logged and turns tracking off for the rest of the fit, so an unreachable server or a rejected metric cannot make models fail, and every run that was started is ended exactly once.

Source code in lazypredict/integrations/mlflow.py

def __init__(self, mlflow_module: Any = None) -> None:
    self._mlflow = mlflow_module
    self.enabled = mlflow_module is not None

call

call(function: Callable[..., Any], *args: Any) -> bool

Run one tracking call; False if tracking is off or the call failed.

Source code in lazypredict/integrations/mlflow.py

def call(self, function: Callable[..., Any], *args: Any) -> bool:
    """Run one tracking call; False if tracking is off or the call failed."""
    if not self.enabled:
        return False
    try:
        function(*args)
    except Exception as exc:
        logger.warning("MLflow tracking failed and is turned off for this fit(): %s", exc)
        self.enabled = False
        return False
    return True

start

start(run_name: str, **params: Any) -> bool

Start a nested run and log params; True if a run was started.

Source code in lazypredict/integrations/mlflow.py

def start(self, run_name: str, **params: Any) -> bool:
    """Start a nested run and log *params*; True if a run was started."""
    if not self.call(start_run, self._mlflow, run_name):
        return False
    for key, value in params.items():
        self.call(self._mlflow.log_param, key, value)
    return True

end

end() -> None

End the run start() started, even if tracking was turned off since.

Source code in lazypredict/integrations/mlflow.py

def end(self) -> None:
    """End the run ``start()`` started, even if tracking was turned off since."""
    try:
        self._mlflow.end_run()
    except Exception as exc:
        logger.warning("Ending the MLflow run failed: %s", exc)

is_mlflow_tracking_enabled

is_mlflow_tracking_enabled() -> bool

Check if MLflow tracking is enabled via the MLFLOW_TRACKING_URI environment variable.

Source code in lazypredict/integrations/mlflow.py

def is_mlflow_tracking_enabled() -> bool:
    """Check if MLflow tracking is enabled via the MLFLOW_TRACKING_URI environment variable."""
    tracking_uri = os.environ.get("MLFLOW_TRACKING_URI")
    return MLFLOW_AVAILABLE and tracking_uri is not None

setup_mlflow

setup_mlflow() -> bool

Point MLflow at MLFLOW_TRACKING_URI when that variable is set.

Autologging is deliberately not enabled: it is a process-wide setting that would change how the user's own code is logged.

Returns:

Type	Description
bool	True if MLflow was successfully configured, False otherwise.

Source code in lazypredict/integrations/mlflow.py

def setup_mlflow() -> bool:
    """Point MLflow at ``MLFLOW_TRACKING_URI`` when that variable is set.

    Autologging is deliberately not enabled: it is a process-wide setting
    that would change how the user's own code is logged.

    Returns
    -------
    bool
        True if MLflow was successfully configured, False otherwise.
    """
    if is_mlflow_tracking_enabled():
        mlflow.set_tracking_uri(os.environ["MLFLOW_TRACKING_URI"])
        return True
    return False

start_run

start_run(mlflow_module: Any, run_name: str) -> Any

Start a run for one model, nested under the caller's run if one is active.

Source code in lazypredict/integrations/mlflow.py

def start_run(mlflow_module: Any, run_name: str) -> Any:
    """Start a run for one model, nested under the caller's run if one is active."""
    return mlflow_module.start_run(run_name=run_name, nested=True)

log_sklearn_model

log_sklearn_model(mlflow_module: Any, model: Any, artifact_name: str, signature: Optional[Any] = None) -> None

Log model as an artifact of the active run without registering it.

MLflow 3 names the artifact with name=; MLflow 2 only accepts artifact_path=.

Source code in lazypredict/integrations/mlflow.py

def log_sklearn_model(
    mlflow_module: Any, model: Any, artifact_name: str, signature: Optional[Any] = None
) -> None:
    """Log *model* as an artifact of the active run without registering it.

    MLflow 3 names the artifact with ``name=``; MLflow 2 only accepts
    ``artifact_path=``.
    """
    log_model = mlflow_module.sklearn.log_model
    key = "name" if "name" in inspect.signature(log_model).parameters else "artifact_path"
    log_model(model, **{key: artifact_name}, signature=signature)

Module contents

integrations

Optional integrations for LazyPredict.