Reduce verbosity for ray events

xgboost_ray/main.py

@@ -366,6 +366,8 @@ class RayParams:
             Defaults to 0 (no retries). Set to -1 for unlimited retries.
         checkpoint_frequency (int): How often to save checkpoints. Defaults
             to ``5`` (every 5th iteration).
+        verbose (bool): Whether to output Ray-specific info messages
+            during training/prediction.
     """
     # Actor scheduling
     num_actors: int = 0
@@ -382,6 +384,8 @@ class RayParams:
     # Distributed callbacks
     distributed_callbacks: Optional[List[DistributedCallback]] = None
 
+    verbose: bool = False
+
     def get_tune_resources(self):
         """Return the resources to use for xgboost_ray training with Tune."""
         if self.cpus_per_actor <= 0 or self.num_actors <= 0:
@@ -977,9 +981,10 @@ def handle_actor_failure(actor_id):
         newly_created += 1
 
     alive_actors = sum(1 for a in _training_state.actors if a is not None)
-    logger.info(f"[RayXGBoost] Created {newly_created} new actors "
-                f"({alive_actors} total actors). Waiting until actors "
-                f"are ready for training.")
+    if ray_params.verbose:
+        logger.info(f"[RayXGBoost] Created {newly_created} new actors "
+                    f"({alive_actors} total actors). Waiting until actors "
+                    f"are ready for training.")
 
     # For distributed datasets (e.g. Modin), this will initialize
     # (and fix) the assignment of data shards to actor ranks
@@ -1022,7 +1027,8 @@ def handle_actor_failure(actor_id):
         _get_actor_alive_status(_training_state.actors, handle_actor_failure)
         raise RayActorError from exc
 
-    logger.info("[RayXGBoost] Starting XGBoost training.")
+    if ray_params.verbose:
+        logger.info("[RayXGBoost] Starting XGBoost training.")
 
     # Start Rabit tracker for gradient sharing
     rabit_process, env = _start_rabit_tracker(alive_actors)
@@ -1513,10 +1519,12 @@ def _wrapped(*args, **kwargs):
     train_additional_results["training_time_s"] = total_training_time
     train_additional_results["total_time_s"] = total_time
 
-    logger.info("[RayXGBoost] Finished XGBoost training on training data "
-                "with total N={total_n:,} in {total_time_s:.2f} seconds "
-                "({training_time_s:.2f} pure XGBoost training time).".format(
-                    **train_additional_results))
+    if ray_params.verbose:
+        logger.info(
+            "[RayXGBoost] Finished XGBoost training on training data "
+            "with total N={total_n:,} in {total_time_s:.2f} seconds "
+            "({training_time_s:.2f} pure XGBoost training time).".format(
+                **train_additional_results))
 
     _shutdown(
         actors=actors,
@@ -1553,7 +1561,8 @@ def _predict(model: xgb.Booster, data: RayDMatrix, ray_params: RayParams,
             distributed_callbacks=ray_params.distributed_callbacks)
         for i in range(ray_params.num_actors)
     ]
-    logger.info(f"[RayXGBoost] Created {len(actors)} remote actors.")
+    if ray_params.verbose:
+        logger.info(f"[RayXGBoost] Created {len(actors)} remote actors.")
 
     # Split data across workers
     wait_load = []
@@ -1570,7 +1579,8 @@ def _predict(model: xgb.Booster, data: RayDMatrix, ray_params: RayParams,
     # Put model into object store
     model_ref = ray.put(model)
 
-    logger.info("[RayXGBoost] Starting XGBoost prediction.")
+    if ray_params.verbose:
+        logger.info("[RayXGBoost] Starting XGBoost prediction.")
 
     # Train
     fut = [actor.predict.remote(model_ref, data, **kwargs) for actor in actors]