#!/usr/bin/env python3
# Copyright (c) Facebook, Inc. and its 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, List, Optional, Tuple
from ax.core.types import TCandidateMetadata, TConfig
from ax.models.torch.alebo import ei_or_nei
from ax.models.torch.botorch import BotorchModel
from ax.models.torch.cbo_sac import generate_model_space_decomposition
from ax.models.torch_base import TorchModel
from ax.utils.common.docutils import copy_doc
from ax.utils.common.logger import get_logger
from botorch.fit import fit_gpytorch_model
from botorch.models.contextual import LCEAGP
from botorch.models.gpytorch import GPyTorchModel
from botorch.models.model_list_gp_regression import ModelListGP
from gpytorch.mlls.exact_marginal_log_likelihood import ExactMarginalLogLikelihood
from torch import Tensor
MIN_OBSERVED_NOISE_LEVEL = 1e-7
logger = get_logger(__name__)
[docs]def get_map_model(
train_X: Tensor,
train_Y: Tensor,
train_Yvar: Tensor,
decomposition: Dict[str, List[int]],
train_embedding: bool = True,
cat_feature_dict: Optional[Dict] = None,
embs_feature_dict: Optional[Dict] = None,
embs_dim_list: Optional[List[int]] = None,
context_weight_dict: Optional[Dict] = None,
) -> Tuple[LCEAGP, ExactMarginalLogLikelihood]:
"""Obtain MAP fitting of Latent Context Embedding Additive (LCE-A) GP."""
# assert train_X is non-batched
assert train_X.dim() < 3, "Don't support batch training"
model = LCEAGP(
train_X=train_X,
train_Y=train_Y,
train_Yvar=train_Yvar,
decomposition=decomposition,
train_embedding=train_embedding,
embs_dim_list=embs_dim_list,
cat_feature_dict=cat_feature_dict,
embs_feature_dict=embs_feature_dict,
context_weight_dict=context_weight_dict,
)
mll = ExactMarginalLogLikelihood(model.likelihood, model)
fit_gpytorch_model(mll)
return model, mll
[docs]class LCEABO(BotorchModel):
r"""Does Bayesian optimization with Latent Context Embedding Additive (LCE-A) GP.
The parameter space decomposition must be provided.
Args:
decomposition: Keys are context names. Values are the lists of parameter
names belong to the context, e.g.
{'context1': ['p1_c1', 'p2_c1'],'context2': ['p1_c2', 'p2_c2']}.
gp_model_args: Dictionary of kwargs to pass to GP model training.
- train_embedding: Boolen. If true, we will train context embedding;
otherwise, we use pre-trained embeddings from embds_feature_dict only.
Default is True.
"""
def __init__(
self,
decomposition: Dict[str, List[str]],
cat_feature_dict: Optional[Dict] = None,
embs_feature_dict: Optional[Dict] = None,
context_weight_dict: Optional[Dict] = None,
embs_dim_list: Optional[List[int]] = None,
gp_model_args: Optional[Dict[str, Any]] = None,
) -> None:
# add validation for input decomposition
for param_list in list(decomposition.values()):
assert len(param_list) == len(
list(decomposition.values())[0]
), "Each Context should contain same number of parameters"
self.decomposition = decomposition
self.cat_feature_dict = cat_feature_dict
self.embs_feature_dict = embs_feature_dict
self.context_weight_dict = context_weight_dict
self.embs_dim_list = embs_dim_list
self.gp_model_args = gp_model_args if gp_model_args is not None else {}
self.feature_names: List[str] = []
self.train_embedding = self.gp_model_args.get("train_embedding", True)
super().__init__(
model_constructor=self.get_and_fit_model,
acqf_constructor=ei_or_nei, # pyre-ignore
)
[docs] @copy_doc(TorchModel.fit)
def fit(
self,
Xs: List[Tensor],
Ys: List[Tensor],
Yvars: List[Tensor],
bounds: List[Tuple[float, float]],
task_features: List[int],
feature_names: List[str],
metric_names: List[str],
fidelity_features: List[int],
candidate_metadata: Optional[List[List[TCandidateMetadata]]] = None,
) -> None:
if len(feature_names) == 0:
raise ValueError("feature names are required for LCEABO")
self.feature_names = feature_names
super().fit(
Xs=Xs,
Ys=Ys,
Yvars=Yvars,
bounds=bounds,
task_features=task_features,
feature_names=feature_names,
metric_names=metric_names,
fidelity_features=fidelity_features,
)
[docs] @copy_doc(TorchModel.best_point)
def best_point(
self,
bounds: List[Tuple[float, float]],
objective_weights: Tensor,
outcome_constraints: Optional[Tuple[Tensor, Tensor]] = None,
linear_constraints: Optional[Tuple[Tensor, Tensor]] = None,
fixed_features: Optional[Dict[int, float]] = None,
model_gen_options: Optional[TConfig] = None,
target_fidelities: Optional[Dict[int, float]] = None,
) -> Optional[Tensor]:
raise NotImplementedError
[docs] def get_and_fit_model(
self,
Xs: List[Tensor],
Ys: List[Tensor],
Yvars: List[Tensor],
task_features: List[int],
fidelity_features: List[int],
metric_names: List[str],
state_dict: Optional[Dict[str, Tensor]] = None,
fidelity_model_id: Optional[int] = None,
**kwargs: Any,
) -> GPyTorchModel:
"""Get a fitted LCEAGP model for each outcome.
Args:
Xs: X for each outcome.
Ys: Y for each outcome.
Yvars: Noise variance of Y for each outcome.
Returns: Fitted LCEAGP model.
"""
# generate model space decomposition dict
decomp_index = generate_model_space_decomposition(
decomposition=self.decomposition, feature_names=self.feature_names
)
models = []
for i, X in enumerate(Xs):
Yvar = Yvars[i].clamp_min_(MIN_OBSERVED_NOISE_LEVEL)
gp_m, _ = get_map_model(
train_X=X,
train_Y=Ys[i],
train_Yvar=Yvar,
decomposition=decomp_index,
train_embedding=self.train_embedding,
cat_feature_dict=self.cat_feature_dict,
embs_feature_dict=self.embs_feature_dict,
embs_dim_list=self.embs_dim_list,
context_weight_dict=self.context_weight_dict,
)
models.append(gp_m)
if len(models) == 1:
model = models[0]
else:
model = ModelListGP(*models)
model.to(Xs[0])
return model
