gpytorchwrapper.src.models.model_train

Functions

create_model_parameters_string(...)

Helper function to print the model parameters generated by model_parameters()

define_likelihood(likelihood_conf, ...)

Define the likelihood used for training the model

define_model(model_conf, model_class, ...)

define_optimizer(model, optimizer_conf)

Define the optimizer used for training the model

loss_figure(train_loss, iteration[, val_loss])

Plot the change of loss during training

model_parameters(model[, transformed])

Helper function to get the current model parameters and parameter values.

train_model(train_x, train_y, training_conf)

Train the model using the training data

training_loop(train_x, train_y, model, mll, ...)

The training loop for the model
gpytorchwrapper.src.models.model_train.create_model_parameters_string(parameter_names: list[str], parameters: list[Tensor | None]) → str[source]

Helper function to print the model parameters generated by model_parameters()

Parameters:

  • parameter_names (list of string) – The parameter name strings

  • parameters (list of Tensor and/or None) – The parameter values

Returns:

return – The string to print

Return type:

str

gpytorchwrapper.src.models.model_train.define_likelihood(likelihood_conf: LikelihoodConf, likelihood_class: Likelihood, train_x: Tensor)[source]

Define the likelihood used for training the model

Parameters:

  • likelihood_conf (OptimizerConf) – Configuration dataclass that specifies the options for the likelihood

  • likelihood_class (Likelihood) – Likelihood class to instantiate

  • train_x (Tensor) – The training input tensor

Returns:

optimizer – The optimizer for the model

Return type:

object

Raises:

KeyError – If a fixed noise likelihood is selected but the noise level is not specified

gpytorchwrapper.src.models.model_train.define_model(model_conf: ModelConf, model_class: ExactGP, train_x: Tensor, train_y: Tensor, likelihood: Likelihood) → ExactGP[source]
Parameters:

  • model_conf (ModelConf) – Model configuration dataclass

  • model_class (ExactGP) – Model class

  • train_x (Tensor) – The training input tensor

  • train_y (Tensor) – The training output tensor

  • likelihood (Likelihood) – Likelihood object

Returns:

Instantiated model

Return type:

Tensor or Distribution or LinearOperator

gpytorchwrapper.src.models.model_train.define_optimizer(model: ExactGP, optimizer_conf: OptimizerConf) → Optimizer[source]

Define the optimizer used for training the model

Parameters:

  • model (object) – The model to be optimized

  • optimizer_conf (OptimizerConf) – Configuration dataclass that specifies the options for the optimizer

Returns:

optimizer – The optimizer for the model

Return type:

object

gpytorchwrapper.src.models.model_train.loss_figure(train_loss: list[float], iteration: list[int], val_loss: list[float] = None) → None[source]

Plot the change of loss during training

Parameters:

  • loss (list) – List containing the loss values

  • iteration (list) – List containing the iteration values

Return type:

None

gpytorchwrapper.src.models.model_train.model_parameters(model: ExactGP, transformed: bool = True) → tuple[list[str], list[Tensor | None]][source]

Helper function to get the current model parameters and parameter values.

Parameters:

  • model (ExactGP) – The model object

  • transformed (bool, default True) – If the parameters must be transformed from the raw parameters used by the model to human-readable ones.

Returns:

parameter_names, parameters – A tuple containing the parameter names and their values

Return type:

tuple of list of str and list of (Tensor or None)

gpytorchwrapper.src.models.model_train.train_model(train_x: Tensor, train_y: Tensor, training_conf: TrainingConf, test_x: Tensor = None, test_y: Tensor = None) → tuple[ExactGP, GaussianLikelihood | MultitaskGaussianLikelihood, dict[str, list[float]]][source]

Train the model using the training data

Parameters:

  • train_x (torch.Tensor) – The input training data

  • train_y (torch.Tensor) – The output training data

  • training_conf (TrainingConf) – Dictionary containing the training specifications

  • test_x (Tensor) – The input test data

  • test_y (Tensor) – The output test data

Returns:

  • model (ExactGP) – The trained model

  • likelihood (Likelihood) – The likelihood of the trained model

  • parameter_dict (dict) – Dictionary containing the parameter values after training

gpytorchwrapper.src.models.model_train.training_loop(train_x: Tensor, train_y: Tensor, model: ExactGP, mll: MarginalLogLikelihood, optimizer: Optimizer, learning_iterations: int, debug: bool, test_x: Tensor = None, test_y: Tensor = None) → None[source]

The training loop for the model

Parameters:

  • train_x (torch.Tensor) – The input training data

  • train_y (torch.Tensor) – The output training data

  • model (object) – The model to be trained

  • mll (MarginalLogLikelihood) – The marginal likelihood of the model

  • optimizer (object) – The optimizer for the model

  • learning_iterations (int) – The number of iterations to train the model

  • debug (bool) – Whether to use the debug mode in GPyTorch. Turn off if unwanted debug exceptions interfere with training.

  • test_x

  • test_y

Return type:

None