Hyperparameter optimization and tuning

Hyperparameters are the parameters used to control the deep learning model training process in IBM Watson Machine Learning Accelerator, such as learning rate and optimization function parameters.

Hyperparameter tuning in IBM Watson Machine Learning Accelerator provides a mechanism to search hyperparameters for a model in a user-defined search space automatically.

Search algorithms are the engine to propose hyperparameter combinations used by a model for training. View the list of available hyperparameter search algorithms, see Hyperparameter search algorithms. To add additional search algorithms to IBM Watson Machine Learning Accelerator, they can be added as search algorithm plugins.

You can develop your own search algorithm plugin which you can add to WML Accelerator using the cluster management console, or using the RESTful API. To add a search algorithm to WML Accelerator using the APIs, see Adding hyperparameter search plugins.

In order to use hyperparameter tuning using the cluster management console, the models must be configured accordingly, see: Hyperparameter tuning with model definition.