snorkel.classification.CheckpointerConfig¶
-
class
snorkel.classification.
CheckpointerConfig
[source]¶ Bases:
tuple
Manager for checkpointing model.
- Parameters
checkpoint_dir – The path to a directory where checkpoints will be saved The Trainer will set this to the log directory if it is None
checkpoint_factor – Check for a best model every this many evaluations. For example, if evaluation_freq is 0.5 epochs and checkpoint_factor is 2, then checkpointing will be attempted every 1 epochs.
checkpoint_metric – The metric to checkpoint on, of the form “task/dataset/split/metric:mode” where mode is “min” or “max”.
checkpoint_task_metrics – Additional metrics to save best models for. Note that the best model according to checkpoint_metric will be the one that is loaded after training and used for early stopping.
checkpoint_runway – No checkpointing will occur for the first this many checkpoint_units
checkpoint_clear – If True, clear all checkpoints besides the best so far.
Methods
Initialize self.
count
(value)index
(value, [start, [stop]])Raises ValueError if the value is not present.
Attributes
Alias for field number 5
Alias for field number 0
Alias for field number 1
Alias for field number 2
Alias for field number 4
Alias for field number 3