Description
A list of potential early stopping strategies for model optimization.
An early stopping strategy can be empty (in which case no early stopping is applied), or can be configured through two parameters: min_delta and patience:
min_delta: The minimum absolute change in the monitored tune metric to qualify as an improvement. Needs to be >= 0.0.patience: The number of epochs with no improvement after which training will be stopped. Needs to be > 0.
By default, training is stopped after three consecutive epochs with no improvements of the tune metric on the validation set.
Supported Task Types
- All
Example
early_stopping:
-
- min_delta: 0.0
patience: 3
- min_delta: 0.1
patience: 0
Updated 6 months ago