autoemulate.emulators.ensemble

class Ensemble(emulators=None, jitter=0.0001, device=None)

Bases: GaussianEmulator

Ensemble emulator that aggregates multiple Emulator instances to provide UQ.

Ensemble emulator that aggregates multiple Emulator instances and returns a GaussianLike representing the ensemble posterior. Note that an Emulator instance may also be an Ensemble itself.

static is_multioutput()

Ensemble supports multi-output.

static get_tune_params()

Return a dictionary of hyperparameters to tune.

class EnsembleMLP(x, y, standardize_x=True, standardize_y=True, activation_cls=<class 'torch.nn.modules.activation.ReLU'>, loss_fn_cls=<class 'torch.nn.modules.loss.MSELoss'>, epochs=100, batch_size=16, layer_dims=None, weight_init='default', scale=1.0, bias_init='default', dropout_prob=None, lr=0.01, params_size=1, random_seed=None, device=None, scheduler_cls=None, scheduler_params=None, n_emulators=4)

Bases: Ensemble

Ensemble of MLP emulators.

This class is an ensemble of MLP emulators, each initialized with the same input and output data.

static is_multioutput()

Ensemble of MLPs supports multi-output.

static get_tune_params()

Return a dictionary of hyperparameters to tune.

class DropoutEnsemble(model, standardize_x=True, standardize_y=True, n_samples=20, jitter=0.0001, device=None)

Bases: GaussianEmulator, TorchDeviceMixin

Monte-Carlo Dropout ensemble.

DropoutEnsemble does a number of forward passes with dropout on, and computes mean and epistemic covariance across them.

static is_multioutput()

DropoutEnsemble supports multi-output.

static get_tune_params()

Return a dictionary of hyperparameters to tune.

class EnsembleMLPDropout(x, y, standardize_x=True, standardize_y=True, activation_cls=<class 'torch.nn.modules.activation.ReLU'>, loss_fn_cls=<class 'torch.nn.modules.loss.MSELoss'>, epochs=100, batch_size=16, layer_dims=None, weight_init='default', scale=1.0, bias_init='default', dropout_prob=0.2, lr=0.01, params_size=1, random_seed=None, device=None, scheduler_cls=None, scheduler_params=None)

Bases: DropoutEnsemble

Ensemble of MLP emulators with dropout.

static get_tune_params()

Return a dictionary of hyperparameters to tune.