snorkel.classification.Checkpointer

class snorkel.classification.Checkpointer(counter_unit, evaluation_freq, **kwargs)[source]

Bases: object

Manager for checkpointing model.

Parameters
  • counter_unit (str) – The unit to use when determining when its time to checkpoint (one of [“epochs”, “batches”, “points”]); must match the counter_unit of LogManager

  • evaluation_freq (float) – How frequently the model is being evaluated (this is the maximum frequency that checkpointing can occur, which will happen if checkpoint_factor==1)

  • kwargs (Any) – Config merged with default_config["checkpointer_config"]

__init__(counter_unit, evaluation_freq, **kwargs)[source]

Initialize self. See help(type(self)) for accurate signature.

Return type

None

Methods

__init__(counter_unit, evaluation_freq, **kwargs)

Initialize self.

checkpoint(iteration, model, metric_dict)

Check if iteration and current metrics necessitate a checkpoint.

clear()

Clear existing checkpoint files, besides the best-yet model.

load_best_model(model)

Load the best model from the checkpoint.

checkpoint(iteration, model, metric_dict)[source]

Check if iteration and current metrics necessitate a checkpoint.

Parameters
  • iteration (float) – Current training iteration

  • model (MultitaskClassifier) – Model to checkpoint

  • metric_dict (Dict[str, float]) – Current performance metrics for model

Return type

None

clear()[source]

Clear existing checkpoint files, besides the best-yet model.

Return type

None

load_best_model(model)[source]

Load the best model from the checkpoint.

Return type

MultitaskClassifier