Source code for ax.storage.botorch_modular_registry

#!/usr/bin/env python3
# Copyright (c) Meta Platforms, Inc. and affiliates.
#
# This source code is licensed under the MIT license found in the
# LICENSE file in the root directory of this source tree.

from typing import Any, Dict, Type

import torch

# Ax `Acquisition` imports
from ax.models.torch.botorch_modular.acquisition import Acquisition
from ax.models.torch.botorch_modular.sebo import SEBOAcquisition

# BoTorch `AcquisitionFunction` imports
from botorch.acquisition.acquisition import AcquisitionFunction

from botorch.acquisition.analytic import (
    ExpectedImprovement,
    LogExpectedImprovement,
    LogNoisyExpectedImprovement,
    NoisyExpectedImprovement,
)
from botorch.acquisition.knowledge_gradient import (
    qKnowledgeGradient,
    qMultiFidelityKnowledgeGradient,
)
from botorch.acquisition.logei import (
    qLogExpectedImprovement,
    qLogNoisyExpectedImprovement,
)
from botorch.acquisition.max_value_entropy_search import (
    qMaxValueEntropy,
    qMultiFidelityMaxValueEntropy,
)
from botorch.acquisition.monte_carlo import (
    qExpectedImprovement,
    qNoisyExpectedImprovement,
)
from botorch.acquisition.multi_objective.logei import (
    qLogExpectedHypervolumeImprovement,
    qLogNoisyExpectedHypervolumeImprovement,
)
from botorch.acquisition.multi_objective.monte_carlo import (
    qExpectedHypervolumeImprovement,
    qNoisyExpectedHypervolumeImprovement,
)
from botorch.models import SaasFullyBayesianSingleTaskGP
from botorch.models.contextual import LCEAGP
from botorch.models.fully_bayesian_multitask import SaasFullyBayesianMultiTaskGP

# BoTorch `Model` imports
from botorch.models.gp_regression import FixedNoiseGP, SingleTaskGP
from botorch.models.gp_regression_fidelity import (
    FixedNoiseMultiFidelityGP,
    SingleTaskMultiFidelityGP,
)
from botorch.models.gp_regression_mixed import MixedSingleTaskGP
from botorch.models.model import Model
from botorch.models.model_list_gp_regression import ModelListGP
from botorch.models.multitask import FixedNoiseMultiTaskGP, MultiTaskGP
from botorch.models.transforms.input import (
    ChainedInputTransform,
    InputPerturbation,
    InputTransform,
    Normalize,
    Round,
    Warp,
)
from botorch.models.transforms.outcome import (
    ChainedOutcomeTransform,
    OutcomeTransform,
    Standardize,
)

# Miscellaneous BoTorch imports
from gpytorch.constraints import Interval
from gpytorch.likelihoods.gaussian_likelihood import GaussianLikelihood
from gpytorch.likelihoods.likelihood import Likelihood

# BoTorch `MarginalLogLikelihood` imports
from gpytorch.mlls.exact_marginal_log_likelihood import ExactMarginalLogLikelihood
from gpytorch.mlls.leave_one_out_pseudo_likelihood import LeaveOneOutPseudoLikelihood
from gpytorch.mlls.marginal_log_likelihood import MarginalLogLikelihood
from gpytorch.mlls.sum_marginal_log_likelihood import SumMarginalLogLikelihood
from gpytorch.priors.torch_priors import GammaPrior

# NOTE: When adding a new registry for a class, make sure to make changes
# to `CLASS_TO_REGISTRY` and `CLASS_TO_REVERSE_REGISTRY` in this file.

"""
Mapping of modular Ax `Acquisition` classes to class name strings.
"""
ACQUISITION_REGISTRY: Dict[Type[Acquisition], str] = {
    Acquisition: "Acquisition",
}


"""
Mapping of BoTorch `Model` classes to class name strings.
"""
MODEL_REGISTRY: Dict[Type[Model], str] = {
    # NOTE: Fixed noise models are deprecated. They point to their
    # supported parent classes, so that we can reap them with minimal
    # concern for backwards compatibility when the time comes.
    FixedNoiseGP: "SingleTaskGP",
    FixedNoiseMultiFidelityGP: "SingleTaskMultiFidelityGP",
    FixedNoiseMultiTaskGP: "MultiTaskGP",
    MixedSingleTaskGP: "MixedSingleTaskGP",
    ModelListGP: "ModelListGP",
    MultiTaskGP: "MultiTaskGP",
    SingleTaskGP: "SingleTaskGP",
    SingleTaskMultiFidelityGP: "SingleTaskMultiFidelityGP",
    SaasFullyBayesianSingleTaskGP: "SaasFullyBayesianSingleTaskGP",
    SaasFullyBayesianMultiTaskGP: "SaasFullyBayesianMultiTaskGP",
    LCEAGP: "LCEAGP",
}


"""
Mapping of Botorch `AcquisitionFunction` classes to class name strings.
"""
ACQUISITION_FUNCTION_REGISTRY: Dict[Type[AcquisitionFunction], str] = {
    ExpectedImprovement: "ExpectedImprovement",
    NoisyExpectedImprovement: "NoisyExpectedImprovement",
    qExpectedHypervolumeImprovement: "qExpectedHypervolumeImprovement",
    qNoisyExpectedHypervolumeImprovement: "qNoisyExpectedHypervolumeImprovement",
    qExpectedImprovement: "qExpectedImprovement",
    qKnowledgeGradient: "qKnowledgeGradient",
    qMaxValueEntropy: "qMaxValueEntropy",
    qMultiFidelityKnowledgeGradient: "qMultiFidelityKnowledgeGradient",
    qMultiFidelityMaxValueEntropy: "qMultiFidelityMaxValueEntropy",
    qNoisyExpectedImprovement: "qNoisyExpectedImprovement",
    # LogEI family below:
    LogExpectedImprovement: "LogExpectedImprovement",
    LogNoisyExpectedImprovement: "LogNoisyExpectedImprovement",
    qLogExpectedImprovement: "qLogExpectedImprovement",
    qLogNoisyExpectedImprovement: "qLogNoisyExpectedImprovement",
    qLogExpectedHypervolumeImprovement: "qLogExpectedHypervolumeImprovement",
    qLogNoisyExpectedHypervolumeImprovement: "qLogNoisyExpectedHypervolumeImprovement",
}


"""
Mapping of BoTorch `MarginalLogLikelihood` classes to class name strings.
"""
MLL_REGISTRY: Dict[Type[MarginalLogLikelihood], str] = {
    ExactMarginalLogLikelihood: "ExactMarginalLogLikelihood",
    LeaveOneOutPseudoLikelihood: "LeaveOneOutPseudoLikelihood",
    SumMarginalLogLikelihood: "SumMarginalLogLikelihood",
}

LIKELIHOOD_REGISTRY: Dict[Type[GaussianLikelihood], str] = {
    GaussianLikelihood: "GaussianLikelihood"
}

GPYTORCH_COMPONENT_REGISTRY: Dict[Type[torch.nn.Module], str] = {
    Interval: "Interval",
    GammaPrior: "GammaPrior",
}

"""
Mapping of BoTorch `InputTransform` classes to class name strings.
"""
INPUT_TRANSFORM_REGISTRY: Dict[Type[InputTransform], str] = {
    ChainedInputTransform: "ChainedInputTransform",
    Normalize: "Normalize",
    Round: "Round",
    Warp: "Warp",
    InputPerturbation: "InputPerturbation",
}

"""
Mapping of BoTorch `OutcomeTransform` classes to class name strings.
"""
OUTCOME_TRANSFORM_REGISTRY: Dict[Type[OutcomeTransform], str] = {
    ChainedOutcomeTransform: "ChainedOutcomeTransform",
    Standardize: "Standardize",
}

"""
Overarching mapping from encoded classes to registry map.
"""
# pyre-fixme[5]: Global annotation cannot contain `Any`.
CLASS_TO_REGISTRY: Dict[Any, Dict[Type[Any], str]] = {
    Acquisition: ACQUISITION_REGISTRY,
    AcquisitionFunction: ACQUISITION_FUNCTION_REGISTRY,
    Likelihood: LIKELIHOOD_REGISTRY,
    MarginalLogLikelihood: MLL_REGISTRY,
    Model: MODEL_REGISTRY,
    Interval: GPYTORCH_COMPONENT_REGISTRY,
    GammaPrior: GPYTORCH_COMPONENT_REGISTRY,
    InputTransform: INPUT_TRANSFORM_REGISTRY,
    OutcomeTransform: OUTCOME_TRANSFORM_REGISTRY,
}


"""
Reverse registries for decoding.
"""
REVERSE_ACQUISITION_REGISTRY: Dict[str, Type[Acquisition]] = {
    v: k for k, v in ACQUISITION_REGISTRY.items()
}


REVERSE_MODEL_REGISTRY: Dict[str, Type[Model]] = {
    "FixedNoiseGP": SingleTaskGP,
    "FixedNoiseMultiFidelityGP": SingleTaskMultiFidelityGP,
    "FixedNoiseMultiTaskGP": MultiTaskGP,
    **{v: k for k, v in MODEL_REGISTRY.items()},
}


REVERSE_ACQUISITION_FUNCTION_REGISTRY: Dict[str, Type[AcquisitionFunction]] = {
    v: k for k, v in ACQUISITION_FUNCTION_REGISTRY.items()
}


REVERSE_MLL_REGISTRY: Dict[str, Type[MarginalLogLikelihood]] = {
    v: k for k, v in MLL_REGISTRY.items()
}


REVERSE_LIKELIHOOD_REGISTRY: Dict[str, Type[Likelihood]] = {
    v: k for k, v in LIKELIHOOD_REGISTRY.items()
}


REVERSE_GPYTORCH_COMPONENT_REGISTRY: Dict[str, Type[torch.nn.Module]] = {
    v: k for k, v in GPYTORCH_COMPONENT_REGISTRY.items()
}


REVERSE_INPUT_TRANSFORM_REGISTRY: Dict[str, Type[InputTransform]] = {
    v: k for k, v in INPUT_TRANSFORM_REGISTRY.items()
}

REVERSE_OUTCOME_TRANSFORM_REGISTRY: Dict[str, Type[OutcomeTransform]] = {
    v: k for k, v in OUTCOME_TRANSFORM_REGISTRY.items()
}

"""
Overarching mapping from encoded classes to reverse registry map.
"""
# pyre-fixme[5]: Global annotation cannot contain `Any`.
CLASS_TO_REVERSE_REGISTRY: Dict[Any, Dict[str, Type[Any]]] = {
    Acquisition: REVERSE_ACQUISITION_REGISTRY,
    AcquisitionFunction: REVERSE_ACQUISITION_FUNCTION_REGISTRY,
    Likelihood: REVERSE_LIKELIHOOD_REGISTRY,
    MarginalLogLikelihood: REVERSE_MLL_REGISTRY,
    Model: REVERSE_MODEL_REGISTRY,
    Interval: REVERSE_GPYTORCH_COMPONENT_REGISTRY,
    GammaPrior: REVERSE_GPYTORCH_COMPONENT_REGISTRY,
    InputTransform: REVERSE_INPUT_TRANSFORM_REGISTRY,
    OutcomeTransform: REVERSE_OUTCOME_TRANSFORM_REGISTRY,
}


[docs]def register_acquisition(acq_class: Type[Acquisition]) -> None: """Add a custom acquisition class to the SQA and JSON registries.""" class_name = acq_class.__name__ CLASS_TO_REGISTRY[Acquisition].update({acq_class: class_name}) CLASS_TO_REVERSE_REGISTRY[Acquisition].update({class_name: acq_class})
[docs]def register_acquisition_function(acqf_class: Type[AcquisitionFunction]) -> None: """Add a custom acquisition class to the SQA and JSON registries.""" class_name = acqf_class.__name__ CLASS_TO_REGISTRY[AcquisitionFunction].update({acqf_class: class_name}) CLASS_TO_REVERSE_REGISTRY[AcquisitionFunction].update({class_name: acqf_class})
[docs]def register_model(model_class: Type[Model]) -> None: """Add a custom model class to the SQA and JSON registries.""" class_name = model_class.__name__ CLASS_TO_REGISTRY[Model].update({model_class: class_name}) CLASS_TO_REVERSE_REGISTRY[Model].update({class_name: model_class})
register_acquisition(SEBOAcquisition)