teradataml RandomSearch provides the capability of early stopping hyperparameter tuning based on the following:
- Time based: Hyperparameter tuning is stopped once the maximum time is reached, thereby ceasing the optimization of hyperparameters.
- Metrics based: Hyperparameter tuning is terminated once a trained model satisfies the specified minimum or maximum thresholds for the respective performance metrics, as dictated by the evaluation criteria.Metrics based method cannot be used for Non-Model Trainer function.
Both time and metrics methods can be used simultaneously, and hyperparameter tuning stops when either of these two methods satisfies the stopping condition.