ModelKernelTrainer¶

class ModelKernelTrainer¶

Bases: TrainerBase

A class for training a polynomial- and kernel-based surrogate model.

Methods

`__init__`	Constructor.
`init_seed`	Initialize with a random seed.
`set_model_class`	Set a model class.
`set_model_params`	Set model parameters to initialize the model class with (`TorchFM` by default).
`set_train_params`	Set training parameters.
`train`	Train a polynomial- and kernel-based model.

__init__(model_class: type[~amplify_bbopt.model.ModelKernel] = <class 'amplify_bbopt.model.ModelKernel'>) → None¶

Constructor.

Parameters:: model_class (Type[ModelKernel], optional) – A model class to train. By calling the constructor, default training parameters are set with ModelKernelTrainer.set_train_params. Defaults to ModelKernel.

__str__() → str¶

Return human-readable training information.

init_seed(seed: int) → None¶

Initialize with a random seed.

set_model_class(model_class: type[~amplify_bbopt.model.ModelKernel] = <class 'amplify_bbopt.model.ModelKernel'>) → None¶

Set a model class.

Parameters:: model_class (Type[ModelKernel]) – A model class. Defaults to ModelKernel.

set_model_params(**model_params: dict) → None¶

Set model parameters to initialize the model class with (TorchFM by default).

The following parameters can be set for the default model class:

beta (float | list[float], optional): A weight for ‘sigma’ in the lower confidence bound (LCB). If multiple values of beta are given as a list, a value in the list is chosen and used in a cyclic manner each time the model is constructed (=in each optimization cycle). When beta = 0 or min(beta) = 0, ModelKernel.is_sigma_required = False, meaning the model uncertainty is not considered. Defaults to 0.0.
gamma (float, optional): A constant in the kernel function to define a linear term in the model. gamma = 0 means no linear term is considered in the model. Note that when the a priori encoding methods (()[conversion.ipynb]) are chosen for all decision variables, linear terms are expected to be modeled naturally in the quadratic terms. Defaults to 0.0.

set_train_params(reg_param: float = 1) → None¶

Set training parameters.

Parameters:: reg_param (float, optional) – A regularization parameter. Defaults to 1.

train( x_values: list[list[bool | int | float]], y_values: list[int | float], logger: Logger | None, ) → ModelKernel¶

Train a polynomial- and kernel-based model.

Parameters:

x_values (list[list[bool | int | float]]) – A list of the input value vectors in the training data (with the last element being a newly added input value vector since the last training).
y_values (list[int | float]) – A list of the corresponding output values in the training data.
logger (Logger | None) – A logger.

Raises:

RuntimeError – If a regularization parameter is not set.

Returns:

A trained model.

Return type:

ModelKernel

__dict__ = mappingproxy({'__module__': 'amplify_bbopt.trainer', '__doc__': 'A class for training a polynomial- and kernel-based surrogate model.', '__init__': <function ModelKernelTrainer.__init__>, 'train': <function ModelKernelTrainer.train>, '_corrcoef': <function ModelKernelTrainer._corrcoef>, 'set_model_class': <function ModelKernelTrainer.set_model_class>, 'set_model_params': <function ModelKernelTrainer.set_model_params>, 'set_train_params': <function ModelKernelTrainer.set_train_params>, '__str__': <function ModelKernelTrainer.__str__>, '__abstractmethods__': frozenset(), '_abc_impl': <_abc._abc_data object>, '__annotations__': {'_model_params': 'dict[str, Any]', '_reg_param': 'float | None', '_beta': 'float | list[float] | None', '_beta_array': 'np.ndarray | None', '_matrix_handler': 'PolyCoefMatrixHandler | None'}})¶