optuna.integration.AllenNLPPruningCallback

class optuna.integration.AllenNLPPruningCallback(trial=None, monitor=None)[source]

AllenNLP callback to prune unpromising trials.

See the example if you want to add a pruning callback which observes a metric.

You can also see the tutorial of our AllenNLP integration on AllenNLP Guide.

Note

When AllenNLPPruningCallback is instantiated in Python script, trial and monitor are mandatory.

On the other hand, when AllenNLPPruningCallback is used with AllenNLPExecutor, trial and monitor would be None. AllenNLPExecutor sets environment variables for a study name, trial id, monitor, and storage. Then AllenNLPPruningCallback loads them to restore trial and monitor.

Parameters
  • trial – A Trial corresponding to the current evaluation of the objective function.

  • monitor – An evaluation metric for pruning, e.g. validation_loss or validation_accuracy.

Note

Added in v2.0.0 as an experimental feature. The interface may change in newer versions without prior notice. See https://github.com/optuna/optuna/releases/tag/v2.0.0.

Methods

on_epoch(trainer, metrics, epoch[, is_primary])

Check if a training reaches saturation.

register(*args, **kwargs)

Stub method for TrainerCallback.register.

on_epoch(trainer, metrics, epoch, is_primary=True, **kwargs)[source]

Check if a training reaches saturation.

Parameters
  • trainer (GradientDescentTrainer) – AllenNLP’s trainer

  • metrics (Dict[str, Any]) – Dictionary of metrics.

  • epoch (int) – Number of current epoch.

  • is_primary (bool) – A flag for AllenNLP internal.

  • kwargs (Any) –

Return type

None

classmethod register(*args, **kwargs)

Stub method for TrainerCallback.register.

This method has the same signature as Registrable.register in AllenNLP.

Parameters
  • args (Any) –

  • kwargs (Any) –

Return type

Callable