Callbacks

Control training with callbacks for early stopping, logging, checkpointing, and more.

NaturalBoost, FormulaBoost, WeibullAFT, and the mean-regression models all accept callbacks= and eval_set= on fit(). FormulaBoost eval tuples are (X, y, model_input); WeibullAFT tuples are (X, y[, event]).

Available Callbacks

Callback	Purpose
EarlyStopping	Stop when validation metric stops improving
Logger	Print training progress
ModelCheckpoint	Save best models during training
LearningRateScheduler	Dynamic learning rate

Early Stopping

import openboost as ob
from openboost import EarlyStopping

model = ob.GradientBoosting(n_trees=500)

model.fit(
    X_train, y_train,
    eval_set=[(X_val, y_val)],
    callbacks=[EarlyStopping(patience=10, min_delta=0.001)],
)

print(f"Stopped at {len(model.trees_)} trees")

Parameters

Parameter	Type	Default	Description
patience	int	10	Rounds without improvement before stopping
min_delta	float	0.0	Minimum change to qualify as improvement
restore_best	bool	True	Restore weights from best iteration

Logger

from openboost import Logger

model.fit(
    X_train, y_train,
    callbacks=[Logger(period=10)],  # Print every 10 trees
)

Output:

[10] train_loss: 0.1234
[20] train_loss: 0.0987
[30] train_loss: 0.0856
...

Model Checkpoint

from openboost import ModelCheckpoint

model.fit(
    X_train, y_train,
    eval_set=[(X_val, y_val)],
    callbacks=[
        ModelCheckpoint(
            filepath='best_model.joblib',
            save_best_only=True,
        )
    ],
)

Learning Rate Scheduler

from openboost import LearningRateScheduler

def lr_schedule(round_num, current_lr):
    """Reduce LR by 10% every 100 rounds."""
    if round_num > 0 and round_num % 100 == 0:
        return current_lr * 0.9
    return current_lr

model.fit(
    X_train, y_train,
    callbacks=[LearningRateScheduler(lr_schedule)],
)

Combining Callbacks

from openboost import EarlyStopping, Logger, ModelCheckpoint

callbacks = [
    EarlyStopping(patience=20),
    Logger(period=10),
    ModelCheckpoint('checkpoints/model_{round}.joblib'),
]

model.fit(
    X_train, y_train,
    eval_set=[(X_val, y_val)],
    callbacks=callbacks,
)

Custom Callbacks

from openboost import Callback

class MyCallback(Callback):
    def on_train_begin(self, state):
        print("Training started!")

    def on_round_end(self, state):
        if state.round % 50 == 0:
            print(f"Round {state.round}: loss = {state.train_loss:.4f}")

    def on_train_end(self, state):
        print(f"Training finished after {state.round} rounds")

model.fit(X_train, y_train, callbacks=[MyCallback()])

Callback Methods

Method	When Called
on_train_begin(state)	Before training starts
on_round_begin(state)	Before each boosting round
on_round_end(state)	After each boosting round
on_train_end(state)	After training completes