This PR adds a callback interface to RayXGBoostActor. This has been a request in #49 and we'll use it for more fine-grained fault tolerance testing - we can e.g. control when actors are allowed to come back after they terminated.
The parameter and class are called "distributed callback" to set them apart from the regular xgboost training callbacks.
This PR adds a callback interface to RayXGBoostActor. This has been a request in #49 and we'll use it for more fine-grained fault tolerance testing - we can e.g. control when actors are allowed to come back after they terminated.
The parameter and class are called "distributed callback" to set them apart from the regular xgboost training callbacks.
Closes #49