Callbacks#

class legateboost.TrainingCallback#

Interface for training callback.

after_iteration(model: LBBase, epoch: int, evals_result: dict[str, dict[str, list[float]]]) → bool#

Run after each iteration. Returns True when training should stop.

Parameters:

  • model – Either a Booster object or a CVPack if the cv function in xgboost is being used.

  • epoch – The current training iteration.

  • evals_result –

    A dictionary containing the evaluation history:

    {"data_name": {"metric_name": [0.5, ...]}}

after_training(model: LBBase) → None#

Run after training is finished.

before_iteration(model: LBBase, epoch: int, evals_result: dict[str, dict[str, list[float]]]) → bool#

Run before each iteration.

Returns True when training should stop. See after_iteration() for details.

before_training(model: LBBase) → None#

Run before training starts.

class legateboost.EarlyStopping(rounds: int, metric_name: str | None = None, min_delta: float = 0.0, prune_model: bool = True, verbose: bool = False)#

Callback for early stopping during training. The last evaluation dataset is used for early stopping.

Parameters:

  • rounds (int) – The number of rounds to wait for improvement before stopping.

  • metric_name (Optional[str]) – The name of the metric to monitor for improvement. If not provided, the last metric in the evaluation result will be used.

  • min_delta (float) – The minimum change in the monitored metric to be considered as improvement.

  • prune_model (bool) – Whether to prune the model after early stopping. If True, the model will be pruned to the best iteration.

Raises:

ValueError – If min_delta is less than 0.

after_iteration(model: LBBase, epoch: int, evals_result: dict[str, dict[str, list[float]]]) → bool#

Run after each iteration. Returns True when training should stop.

Parameters:

  • model – Either a Booster object or a CVPack if the cv function in xgboost is being used.

  • epoch – The current training iteration.

  • evals_result –

    A dictionary containing the evaluation history:

    {"data_name": {"metric_name": [0.5, ...]}}

after_training(model: LBBase) → None#

Run after training is finished.

before_training(model: LBBase) → None#

Run before training starts.