objective_naswot

Objective functions for hyperparameter optimizations without training.

class ObjectiveNASWOT(search_parameters: Parameters, data_handler: DataHandler, trial_type, batch_size=None)[source]

Bases: ObjectiveBase

Represents the objective function using no NN training.

The objective value is calculated using the Jacobian.

Parameters:

  • search_parameters (mala.common.parameters.Parameters) – Parameters used to create this objective.

  • data_handler (mala.datahandling.data_handler.DataHandler) – datahandler to be used during the hyperparameter optimization.

  • trial_type (string) –

    Format of hyperparameters used in this objective. Supported choices are:

    • optuna

    • oat

  • batch_size (int) – Batch size for the forwarding. Default (None) means that the regular mini batch size from ParametersRunning is used; however, for some applications it might make sense to specify something different.