Callbacks

Training callbacks for logging, checkpointing, and early stopping.

See Also

• Monitoring Overview - All monitoring tools
• Metrics - Metric tracking
• Checkpointing - Save/restore state

---

datarax.monitoring.callbacks

Callback system for Datarax monitoring.

This module implements the observer pattern for metrics collection and event handling in Datarax pipelines.

logger module-attribute

logger = getLogger(__name__)

MetricsObserver

Base class for metrics observers.

Metrics observers receive notifications about metrics collected during pipeline operation.

update

update(metrics: list[MetricRecord]) -> None

Handle updated metrics.

Parameters:

Name	Type	Description	Default
metrics	list[MetricRecord]	List of new metric records.	required

reset

reset() -> None

Reset observer-local state before a new monitoring run.

CallbackRegistry

CallbackRegistry()

Registry for callbacks during pipeline operation.

This class manages a collection of observers that receive notifications about metrics and events.

register

register(observer: MetricsObserver) -> None

Register an observer.

Parameters:

Name	Type	Description	Default
observer	MetricsObserver	Observer to register.	required

is_observer_unregistered

is_observer_unregistered(observer: MetricsObserver) -> bool

Unregister an observer.

Parameters:

Name	Type	Description	Default
observer	MetricsObserver	Observer to is_observer_unregistered.	required

Returns:

Type	Description
bool	True if the observer was found and removed, False otherwise.

notify

notify(metrics: list[MetricRecord]) -> None

Notify all observers of new metrics.

Parameters:

Name	Type	Description	Default
metrics	list[MetricRecord]	List of new metric records.	required

clear

clear() -> None

Remove all registered observers.