#!/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.
# pyre-strict
import json
import time
import warnings
from abc import ABC
from collections import OrderedDict
from copy import deepcopy
from dataclasses import dataclass, field
from logging import Logger
from typing import Any, Dict, List, MutableMapping, Optional, Set, Tuple, Type
from ax.core.arm import Arm
from ax.core.base_trial import NON_ABANDONED_STATUSES, TrialStatus
from ax.core.data import Data
from ax.core.experiment import Experiment
from ax.core.generator_run import extract_arm_predictions, GeneratorRun
from ax.core.observation import (
Observation,
ObservationData,
ObservationFeatures,
observations_from_data,
recombine_observations,
separate_observations,
)
from ax.core.optimization_config import OptimizationConfig
from ax.core.parameter import ParameterType, RangeParameter
from ax.core.search_space import SearchSpace
from ax.core.types import TCandidateMetadata, TModelCov, TModelMean, TModelPredict
from ax.exceptions.core import UnsupportedError, UserInputError
from ax.modelbridge.transforms.base import Transform
from ax.modelbridge.transforms.cast import Cast
from ax.models.types import TConfig
from ax.utils.common.logger import get_logger
from ax.utils.common.typeutils import checked_cast, not_none
from botorch.exceptions.warnings import InputDataWarning
logger: Logger = get_logger(__name__)
[docs]@dataclass(frozen=True)
class BaseGenArgs:
search_space: SearchSpace
optimization_config: Optional[OptimizationConfig]
pending_observations: Dict[str, List[ObservationFeatures]]
fixed_features: Optional[ObservationFeatures]
[docs]@dataclass(frozen=True)
class GenResults:
observation_features: List[ObservationFeatures]
weights: List[float]
best_observation_features: Optional[ObservationFeatures] = None
gen_metadata: Dict[str, Any] = field(default_factory=dict)
[docs]class ModelBridge(ABC): # noqa: B024 -- ModelBridge doesn't have any abstract methods.
"""The main object for using models in Ax.
ModelBridge specifies 3 methods for using models:
- predict: Make model predictions. This method is not optimized for
speed and so should be used primarily for plotting or similar tasks
and not inside an optimization loop.
- gen: Use the model to generate new candidates.
- cross_validate: Do cross validation to assess model predictions.
ModelBridge converts Ax types like Data and Arm to types that are
meant to be consumed by the models. The data sent to the model will depend
on the implementation of the subclass, which will specify the actual API
for external model.
This class also applies a sequence of transforms to the input data and
problem specification which can be used to ensure that the external model
receives appropriate inputs.
Subclasses will implement what is here referred to as the "terminal
transform," which is a transform that changes types of the data and problem
specification.
"""
def __init__(
self,
search_space: SearchSpace,
# pyre-fixme[2]: Parameter annotation cannot be `Any`.
model: Any,
transforms: Optional[List[Type[Transform]]] = None,
experiment: Optional[Experiment] = None,
data: Optional[Data] = None,
transform_configs: Optional[Dict[str, TConfig]] = None,
status_quo_name: Optional[str] = None,
status_quo_features: Optional[ObservationFeatures] = None,
optimization_config: Optional[OptimizationConfig] = None,
fit_out_of_design: bool = False,
fit_abandoned: bool = False,
fit_tracking_metrics: bool = True,
fit_on_init: bool = True,
) -> None:
"""
Applies transforms and fits model.
Args:
experiment: Is used to get arm parameters. Is not mutated.
search_space: Search space for fitting the model. Constraints need
not be the same ones used in gen.
data: Ax Data.
model: Interface will be specified in subclass. If model requires
initialization, that should be done prior to its use here.
transforms: List of uninitialized transform classes. Forward
transforms will be applied in this order, and untransforms in
the reverse order.
transform_configs: A dictionary from transform name to the
transform config dictionary.
status_quo_name: Name of the status quo arm. Can only be used if
Data has a single set of ObservationFeatures corresponding to
that arm.
status_quo_features: ObservationFeatures to use as status quo.
Either this or status_quo_name should be specified, not both.
optimization_config: Optimization config defining how to optimize
the model.
fit_out_of_design: If specified, all training data is returned.
Otherwise, only in design points are returned.
fit_abandoned: Whether data for abandoned arms or trials should be
included in model training data. If ``False``, only
non-abandoned points are returned.
fit_tracking_metrics: Whether to fit a model for tracking metrics.
Setting this to False will improve runtime at the expense of
models not being available for predicting tracking metrics.
NOTE: This can only be set to False when the optimization config
is provided.
fit_on_init: Whether to fit the model on initialization. This can
be used to skip model fitting when a fitted model is not needed.
To fit the model afterwards, use `_process_and_transform_data`
to get the transformed inputs and call `_fit_if_implemented` with
the transformed inputs.
"""
t_fit_start = time.monotonic()
transforms = transforms or []
# pyre-ignore: Cast is a Tranform
transforms: List[Type[Transform]] = [Cast] + transforms
self.fit_time: float = 0.0
self.fit_time_since_gen: float = 0.0
self._metric_names: Set[str] = set()
self._training_data: List[Observation] = []
self._optimization_config: Optional[OptimizationConfig] = optimization_config
self._training_in_design: List[bool] = []
self._status_quo: Optional[Observation] = None
self._status_quo_name: Optional[str] = None
self._arms_by_signature: Optional[Dict[str, Arm]] = None
self.transforms: MutableMapping[str, Transform] = OrderedDict()
self._model_key: Optional[str] = None
self._model_kwargs: Optional[Dict[str, Any]] = None
self._bridge_kwargs: Optional[Dict[str, Any]] = None
self._model_space: SearchSpace = search_space.clone()
self._raw_transforms = transforms
self._transform_configs: Optional[Dict[str, TConfig]] = transform_configs
self._fit_out_of_design = fit_out_of_design
self._fit_abandoned = fit_abandoned
self._fit_tracking_metrics = fit_tracking_metrics
self.outcomes: List[str] = []
self._experiment_has_immutable_search_space_and_opt_config: bool = (
experiment is not None and experiment.immutable_search_space_and_opt_config
)
self._experiment_properties: Dict[str, Any] = {}
if experiment is not None:
if self._optimization_config is None:
self._optimization_config = experiment.optimization_config
self._arms_by_signature = experiment.arms_by_signature
self._experiment_properties = experiment._properties
if self._fit_tracking_metrics is False:
if self._optimization_config is None:
raise UserInputError(
"Optimization config is required when "
"`fit_tracking_metrics` is False."
)
self.outcomes = sorted(self._optimization_config.metrics.keys())
# Set training data (in the raw / untransformed space). This also omits
# out-of-design and abandoned observations depending on the corresponding flags.
observations_raw = self._prepare_observations(experiment=experiment, data=data)
observations_raw = self._set_training_data(
observations=observations_raw, search_space=self._model_space
)
# Set model status quo.
# NOTE: training data must be set before setting the status quo.
self._set_status_quo(
experiment=experiment,
status_quo_name=status_quo_name,
status_quo_features=status_quo_features,
)
# Save model, apply terminal transform, and fit.
self.model = model
if fit_on_init:
observations, search_space = self._transform_data(
observations=observations_raw,
search_space=self._model_space,
transforms=self._raw_transforms,
transform_configs=self._transform_configs,
)
self._fit_if_implemented(
search_space=search_space,
observations=observations,
time_so_far=time.monotonic() - t_fit_start,
)
def _fit_if_implemented(
self,
search_space: SearchSpace,
observations: List[Observation],
time_so_far: float,
) -> None:
r"""Fits the model if `_fit` is implemented and stores fit time.
Args:
search_space: A transformed search space for fitting the model.
observations: The observations to fit the model with. These should
also be transformed.
time_so_far: Time spent in initializing the model up to
`_fit_if_implemented` call.
"""
try:
t_fit_start = time.monotonic()
self._fit(
model=self.model,
search_space=search_space,
observations=observations,
)
increment = time.monotonic() - t_fit_start + time_so_far
self.fit_time += increment
self.fit_time_since_gen += increment
except NotImplementedError:
pass
def _process_and_transform_data(
self,
experiment: Optional[Experiment] = None,
data: Optional[Data] = None,
) -> Tuple[List[Observation], SearchSpace]:
r"""Processes the data into observations and returns transformed
observations and the search space. This packages the following methods:
* self._prepare_observations
* self._set_training_data
* self._transform_data
"""
observations = self._prepare_observations(experiment=experiment, data=data)
observations_raw = self._set_training_data(
observations=observations, search_space=self._model_space
)
return self._transform_data(
observations=observations_raw,
search_space=self._model_space,
transforms=self._raw_transforms,
transform_configs=self._transform_configs,
)
def _prepare_observations(
self, experiment: Optional[Experiment], data: Optional[Data]
) -> List[Observation]:
if experiment is None or data is None:
return []
return observations_from_data(
experiment=experiment,
data=data,
statuses_to_include=self.statuses_to_fit,
statuses_to_include_map_metric=self.statuses_to_fit_map_metric,
)
def _transform_data(
self,
observations: List[Observation],
search_space: SearchSpace,
transforms: Optional[List[Type[Transform]]],
transform_configs: Optional[Dict[str, TConfig]],
assign_transforms: bool = True,
) -> Tuple[List[Observation], SearchSpace]:
"""Initialize transforms and apply them to provided data."""
# Initialize transforms
search_space = search_space.clone()
if transforms is not None:
if transform_configs is None:
transform_configs = {}
for t in transforms:
t_instance = t(
search_space=search_space,
observations=observations,
modelbridge=self,
config=transform_configs.get(t.__name__, None),
)
search_space = t_instance.transform_search_space(search_space)
observations = t_instance.transform_observations(observations)
if assign_transforms:
self.transforms[t.__name__] = t_instance
return observations, search_space
def _prepare_training_data(
self, observations: List[Observation]
) -> List[Observation]:
observation_features, observation_data = separate_observations(observations)
if len(observation_features) != len(set(observation_features)):
raise ValueError(
"Observation features are not unique. "
"Something went wrong constructing training data..."
)
return observations
def _set_training_data(
self, observations: List[Observation], search_space: SearchSpace
) -> List[Observation]:
"""Store training data, not-transformed.
If the modelbridge specifies _fit_out_of_design, all training data is
returned. Otherwise, only in design points are returned.
"""
observations = self._prepare_training_data(observations=observations)
self._training_data = deepcopy(observations)
self._metric_names: Set[str] = set()
for obs in observations:
self._metric_names.update(obs.data.metric_names)
return self._process_in_design(
search_space=search_space,
observations=observations,
)
def _extend_training_data(
self, observations: List[Observation]
) -> List[Observation]:
"""Extend and return training data, not-transformed.
If the modelbridge specifies _fit_out_of_design, all training data is
returned. Otherwise, only in design points are returned.
Args:
observations: New observations.
Returns: New + old observations.
"""
observations = self._prepare_training_data(observations=observations)
for obs in observations:
for metric_name in obs.data.metric_names:
if metric_name not in self._metric_names:
raise ValueError(
f"Unrecognised metric {metric_name}; cannot update "
"training data with metrics that were not in the original "
"training data."
)
# Initialize with all points in design.
self._training_data.extend(deepcopy(observations))
all_observations = self.get_training_data()
return self._process_in_design(
search_space=self._model_space,
observations=all_observations,
)
def _process_in_design(
self,
search_space: SearchSpace,
observations: List[Observation],
) -> List[Observation]:
"""Set training_in_design, and decide whether to filter out of design points."""
# Don't filter points.
if self._fit_out_of_design:
# Use all data for training
# Set training_in_design to True for all observations so that
# all observations are used in CV and plotting
self.training_in_design = [True] * len(observations)
return observations
in_design = self._compute_in_design(
search_space=search_space, observations=observations
)
self.training_in_design = in_design
in_design_obs = [
observations[i] for i, is_in_design in enumerate(in_design) if is_in_design
]
return in_design_obs
def _compute_in_design(
self,
search_space: SearchSpace,
observations: List[Observation],
) -> List[bool]:
return [
search_space.check_membership(obs.features.parameters)
for obs in observations
]
def _set_status_quo(
self,
experiment: Optional[Experiment],
status_quo_name: Optional[str],
status_quo_features: Optional[ObservationFeatures],
) -> None:
"""Set model status quo by matching status_quo_name or status_quo_features.
First checks for status quo in inputs status_quo_name and
status_quo_features. If neither of these is provided, checks the
experiment for a status quo. If that is set, it is handled by name in
the same way as input status_quo_name.
Args:
experiment: Experiment that will be checked for status quo.
status_quo_name: Name of status quo arm.
status_quo_features: Features for status quo.
"""
self._status_quo: Optional[Observation] = None
sq_obs = None
if (
status_quo_name is None
and status_quo_features is None
and experiment is not None
and experiment.status_quo is not None
):
status_quo_name = experiment.status_quo.name
if status_quo_name is not None:
if status_quo_features is not None:
raise ValueError(
"Specify either status_quo_name or status_quo_features, not both."
)
sq_obs = [
obs for obs in self._training_data if obs.arm_name == status_quo_name
]
elif status_quo_features is not None:
sq_obs = [
obs
for obs in self._training_data
if (obs.features.parameters == status_quo_features.parameters)
and (obs.features.trial_index == status_quo_features.trial_index)
]
# if status_quo_name or status_quo_features is used for matching status quo
if sq_obs is not None:
if len(sq_obs) == 0:
logger.warning(f"Status quo {status_quo_name} not present in data")
elif len(sq_obs) >= 1:
# status quo name (not features as trial index is part of the
# observation features) should be consistent even if we have multiple
# observations of the status quo.
# This is useful for getting status_quo_data_by_trial
self._status_quo_name = sq_obs[0].arm_name
if len(sq_obs) > 1:
logger.warning(
f"Status quo {status_quo_name} found in data with multiple "
"features. Use status_quo_features to specify which to use."
)
else:
# if there is a unique status_quo, set it
# unique features verified in _set_training_data.
self._status_quo = sq_obs[0]
@property
def status_quo_data_by_trial(self) -> Optional[Dict[int, ObservationData]]:
"""A map of trial index to the status quo observation data of each trial"""
return _get_status_quo_by_trial(
observations=self._training_data,
status_quo_name=(
self._status_quo_name
if self.status_quo is None
else self.status_quo.arm_name
),
status_quo_features=(
None if self.status_quo is None else self.status_quo.features
),
)
@property
def status_quo(self) -> Optional[Observation]:
"""Observation corresponding to status quo, if any."""
return self._status_quo
@property
def metric_names(self) -> Set[str]:
"""Metric names present in training data."""
return self._metric_names
@property
def model_space(self) -> SearchSpace:
"""SearchSpace used to fit model."""
return self._model_space
[docs] def get_training_data(self) -> List[Observation]:
"""A copy of the (untransformed) data with which the model was fit."""
return deepcopy(self._training_data)
@property
def training_in_design(self) -> List[bool]:
"""For each observation in the training data, a bool indicating if it
is in-design for the model.
"""
return self._training_in_design
@property
def statuses_to_fit(self) -> Set[TrialStatus]:
"""Statuses to fit the model on."""
if self._fit_abandoned:
return set(TrialStatus)
return NON_ABANDONED_STATUSES
@property
def statuses_to_fit_map_metric(self) -> Set[TrialStatus]:
"""Statuses to fit the model on."""
return {TrialStatus.COMPLETED}
@training_in_design.setter
def training_in_design(self, training_in_design: List[bool]) -> None:
if len(training_in_design) != len(self._training_data):
raise ValueError(
f"In-design indicators not same length ({len(training_in_design)})"
f" as training data ({len(self._training_data)})."
)
# Identify out-of-design arms
if sum(training_in_design) < len(training_in_design):
ood_names = []
for i, obs in enumerate(self._training_data):
if not training_in_design[i] and obs.arm_name is not None:
ood_names.append(obs.arm_name)
ood_str = ", ".join(set(ood_names))
logger.info(f"Leaving out out-of-design observations for arms: {ood_str}")
self._training_in_design = training_in_design
def _fit(
self,
# pyre-fixme[2]: Parameter annotation cannot be `Any`.
model: Any,
search_space: SearchSpace,
observations: List[Observation],
) -> None:
"""Apply terminal transform and fit model."""
raise NotImplementedError
def _batch_predict(
self, observation_features: List[ObservationFeatures]
) -> List[ObservationData]:
"""Predict a list of ObservationFeatures together."""
# Get modifiable version
observation_features = deepcopy(observation_features)
# Transform
for t in self.transforms.values():
observation_features = t.transform_observation_features(
observation_features
)
# Apply terminal transform and predict
observation_data = self._predict(observation_features)
# Apply reverse transforms, in reverse order
pred_observations = recombine_observations(
observation_features=observation_features, observation_data=observation_data
)
for t in reversed(list(self.transforms.values())):
pred_observations = t.untransform_observations(pred_observations)
return [obs.data for obs in pred_observations]
def _single_predict(
self, observation_features: List[ObservationFeatures]
) -> List[ObservationData]:
"""Predict one ObservationFeature at a time."""
observation_data = []
for obsf in observation_features:
try:
obsd = self._batch_predict([obsf])
observation_data += obsd
except (TypeError, ValueError) as e:
# If the prediction is not out of design, this is a real error.
# Let's re-raise.
if self.model_space.check_membership(obsf.parameters):
logger.debug(obsf.parameters)
logger.debug(self.model_space)
raise e from None
# Prediction is out of design.
# Training data is untranformed already.
observation = next(
(
data
for data in self.get_training_data()
if obsf.parameters == data.features.parameters
and obsf.trial_index == data.features.trial_index
),
None,
)
if not observation:
raise ValueError(
"Out-of-design point could not be transformed, and was "
"not found in the training data."
)
observation_data.append(observation.data)
return observation_data
def _predict_observation_data(
self, observation_features: List[ObservationFeatures]
) -> List[ObservationData]:
"""
Like 'predict' method, but returns results as a list of ObservationData
Predictions are made for all outcomes.
If an out-of-design observation can successfully be transformed,
the predicted value will be returned.
Othwerise, we will attempt to find that observation in the training data
and return the raw value.
Args:
observation_features: observation features
Returns:
List of `ObservationData`
"""
# Predict in single batch.
try:
observation_data = self._batch_predict(observation_features)
# Predict one by one.
except (TypeError, ValueError):
observation_data = self._single_predict(observation_features)
return observation_data
[docs] def predict(self, observation_features: List[ObservationFeatures]) -> TModelPredict:
"""Make model predictions (mean and covariance) for the given
observation features.
Predictions are made for all outcomes.
If an out-of-design observation can successfully be transformed,
the predicted value will be returned.
Othwerise, we will attempt to find that observation in the training data
and return the raw value.
Args:
observation_features: observation features
Returns:
2-element tuple containing
- Dictionary from metric name to list of mean estimates, in same
order as observation_features.
- Nested dictionary with cov['metric1']['metric2'] a list of
cov(metric1@x, metric2@x) for x in observation_features.
"""
observation_data = self._predict_observation_data(
observation_features=observation_features
)
f, cov = unwrap_observation_data(observation_data)
return f, cov
def _predict(
self, observation_features: List[ObservationFeatures]
) -> List[ObservationData]:
"""Apply terminal transform, predict, and reverse terminal transform on
output.
"""
raise NotImplementedError
[docs] def update(self, new_data: Data, experiment: Experiment) -> None:
"""Update the model bridge and the underlying model with new data. This
method should be used instead of `fit`, in cases where the underlying
model does not need to be re-fit from scratch, but rather updated.
Note: `update` expects only new data (obtained since the model initialization
or last update) to be passed in, not all data in the experiment.
Args:
new_data: Data from the experiment obtained since the last call to
`update`.
experiment: Experiment, in which this data was obtained.
"""
raise DeprecationWarning("ModelBridge.update is deprecated. Use `fit` instead.")
def _get_transformed_gen_args(
self,
search_space: SearchSpace,
optimization_config: Optional[OptimizationConfig] = None,
pending_observations: Optional[Dict[str, List[ObservationFeatures]]] = None,
fixed_features: Optional[ObservationFeatures] = None,
) -> BaseGenArgs:
if pending_observations is None:
pending_observations = {}
if optimization_config is None:
optimization_config = (
self._optimization_config.clone()
if self._optimization_config is not None
else None
)
else:
if not self._fit_tracking_metrics:
# Check that the optimization config has the same metrics as
# the original one. Otherwise, we may attempt to optimize over
# metrics that do not have a fitted model.
outcomes = set(optimization_config.metrics.keys())
if not outcomes.issubset(self.outcomes):
raise UnsupportedError(
"When fit_tracking_metrics is False, the optimization config "
"can only include metrics that were included in the "
"optimization config used while initializing the ModelBridge. "
f"Metrics {outcomes} is not a subset of {self.outcomes}."
)
optimization_config = optimization_config.clone()
# TODO(T34225037): replace deepcopy with native clone() in Ax
pending_observations = deepcopy(pending_observations)
fixed_features = deepcopy(fixed_features)
search_space = search_space.clone()
# Transform
for t in self.transforms.values():
search_space = t.transform_search_space(search_space)
if optimization_config is not None:
optimization_config = t.transform_optimization_config(
optimization_config=optimization_config,
modelbridge=self,
fixed_features=fixed_features,
)
for metric, po in pending_observations.items():
pending_observations[metric] = t.transform_observation_features(po)
fixed_features = (
t.transform_observation_features([fixed_features])[0]
if fixed_features is not None
else None
)
return BaseGenArgs(
search_space=search_space,
optimization_config=optimization_config,
pending_observations=pending_observations,
fixed_features=fixed_features,
)
def _validate_gen_inputs(
self,
n: int,
search_space: Optional[SearchSpace] = None,
optimization_config: Optional[OptimizationConfig] = None,
pending_observations: Optional[Dict[str, List[ObservationFeatures]]] = None,
fixed_features: Optional[ObservationFeatures] = None,
model_gen_options: Optional[TConfig] = None,
) -> None:
"""Validate inputs to `ModelBridge.gen`.
Currently, this is only used to ensure that `n` is a positive integer.
"""
if n < 1:
raise UserInputError(
f"Attempted to generate n={n} points. Number of points to generate "
"must be a positive integer."
)
[docs] def gen(
self,
n: int,
search_space: Optional[SearchSpace] = None,
optimization_config: Optional[OptimizationConfig] = None,
pending_observations: Optional[Dict[str, List[ObservationFeatures]]] = None,
fixed_features: Optional[ObservationFeatures] = None,
model_gen_options: Optional[TConfig] = None,
) -> GeneratorRun:
"""
Generate new points from the underlying model according to
search_space, optimization_config and other parameters.
Args:
n: Number of points to generate
search_space: Search space
optimization_config: Optimization config
pending_observations: A map from metric name to pending
observations for that metric.
fixed_features: An ObservationFeatures object containing any
features that should be fixed at specified values during
generation.
model_gen_options: A config dictionary that is passed along to the
model. See `TorchOptConfig` for details.
Returns:
A GeneratorRun object that contains the generated points and other metadata.
"""
self._validate_gen_inputs(
n=n,
search_space=search_space,
optimization_config=optimization_config,
pending_observations=pending_observations,
fixed_features=fixed_features,
model_gen_options=model_gen_options,
)
t_gen_start = time.monotonic()
# Get modifiable versions
if search_space is None:
search_space = self._model_space
orig_search_space = search_space
search_space = search_space.clone()
base_gen_args = self._get_transformed_gen_args(
search_space=search_space,
optimization_config=optimization_config,
pending_observations=pending_observations,
fixed_features=fixed_features,
)
# Apply terminal transform and gen
gen_results = self._gen(
n=n,
search_space=base_gen_args.search_space,
optimization_config=base_gen_args.optimization_config,
pending_observations=base_gen_args.pending_observations,
fixed_features=base_gen_args.fixed_features,
model_gen_options=model_gen_options,
)
observation_features = gen_results.observation_features
best_obsf = gen_results.best_observation_features
# Apply reverse transforms
for t in reversed(list(self.transforms.values())):
observation_features = t.untransform_observation_features(
observation_features
)
if best_obsf is not None:
best_obsf = t.untransform_observation_features([best_obsf])[0]
# Clamp the untransformed data to the original search space if
# we don't fit/gen OOD points
if not self._fit_out_of_design:
observation_features = clamp_observation_features(
observation_features, orig_search_space
)
if best_obsf is not None:
best_obsf = clamp_observation_features([best_obsf], orig_search_space)[
0
]
best_point_predictions = None
try:
model_predictions = self.predict(observation_features)
if best_obsf is not None:
best_point_predictions = extract_arm_predictions(
model_predictions=self.predict([best_obsf]), arm_idx=0
)
except NotImplementedError:
model_predictions = None
if best_obsf is None:
best_arm = None
else:
best_arms, _ = gen_arms(
observation_features=[best_obsf],
arms_by_signature=self._arms_by_signature,
)
best_arm = best_arms[0]
arms, candidate_metadata = gen_arms(
observation_features=observation_features,
arms_by_signature=self._arms_by_signature,
)
# If experiment has immutable search space and metrics, no need to
# save them on generator runs.
immutable = getattr(
self, "_experiment_has_immutable_search_space_and_opt_config", False
)
optimization_config = None if immutable else base_gen_args.optimization_config
gr = GeneratorRun(
arms=arms,
weights=gen_results.weights,
optimization_config=optimization_config,
search_space=None if immutable else base_gen_args.search_space,
model_predictions=model_predictions,
best_arm_predictions=(
None if best_arm is None else (best_arm, best_point_predictions)
),
fit_time=self.fit_time_since_gen,
gen_time=time.monotonic() - t_gen_start,
model_key=self._model_key,
model_kwargs=self._model_kwargs,
bridge_kwargs=self._bridge_kwargs,
gen_metadata=gen_results.gen_metadata,
model_state_after_gen=self._get_serialized_model_state(),
candidate_metadata_by_arm_signature=candidate_metadata,
)
if len(gr.arms) < n:
logger.warning(
f"{self} was not able to generate {n} unique candidates. "
"Generated arms have the following weights, as there are repeats:\n"
f"{gr.weights}"
)
self.fit_time_since_gen = 0.0
return gr
def _gen(
self,
n: int,
search_space: SearchSpace,
optimization_config: Optional[OptimizationConfig],
pending_observations: Dict[str, List[ObservationFeatures]],
fixed_features: Optional[ObservationFeatures],
model_gen_options: Optional[TConfig],
) -> GenResults:
"""Apply terminal transform, gen, and reverse terminal transform on
output.
"""
raise NotImplementedError
[docs] def cross_validate(
self,
cv_training_data: List[Observation],
cv_test_points: List[ObservationFeatures],
use_posterior_predictive: bool = False,
) -> List[ObservationData]:
"""Make a set of cross-validation predictions.
Args:
cv_training_data: The training data to use for cross validation.
cv_test_points: The test points at which predictions will be made.
use_posterior_predictive: A boolean indicating if the predictions
should be from the posterior predictive (i.e. including
observation noise).
Returns:
A list of predictions at the test points.
"""
# Apply transforms to cv_training_data and cv_test_points
cv_training_data, cv_test_points, search_space = self._transform_inputs_for_cv(
cv_training_data=cv_training_data, cv_test_points=cv_test_points
)
# Apply terminal transform, and get predictions.
with warnings.catch_warnings():
# Since each CV fold removes points from the training data, the remaining
# observations will not pass the standardization test. To avoid confusing
# users with this warning, we filter it out.
warnings.filterwarnings(
"ignore",
message="Data is not standardized",
category=InputDataWarning,
)
cv_predictions = self._cross_validate(
search_space=search_space,
cv_training_data=cv_training_data,
cv_test_points=cv_test_points,
use_posterior_predictive=use_posterior_predictive,
)
# Apply reverse transforms, in reverse order
cv_test_observations = [
Observation(features=obsf, data=cv_predictions[i])
for i, obsf in enumerate(cv_test_points)
]
for t in reversed(list(self.transforms.values())):
cv_test_observations = t.untransform_observations(cv_test_observations)
return [obs.data for obs in cv_test_observations]
def _cross_validate(
self,
search_space: SearchSpace,
cv_training_data: List[Observation],
cv_test_points: List[ObservationFeatures],
use_posterior_predictive: bool = False,
) -> List[ObservationData]:
"""Apply the terminal transform, make predictions on the test points,
and reverse terminal transform on the results.
"""
raise NotImplementedError
def _transform_inputs_for_cv(
self,
cv_training_data: List[Observation],
cv_test_points: List[ObservationFeatures],
) -> Tuple[List[Observation], List[ObservationFeatures], SearchSpace]:
"""Apply transforms to cv_training_data and cv_test_points,
and return cv_training_data, cv_test_points, and search space in
transformed space. This is to prepare data to be used in _cross_validate.
Args:
cv_training_data: The training data to use for cross validation.
cv_test_points: The test points at which predictions will be made.
Returns:
cv_training_data, cv_test_points, and search space
in transformed space."""
cv_test_points = deepcopy(cv_test_points)
cv_training_data = deepcopy(cv_training_data)
search_space = self._model_space.clone()
for t in self.transforms.values():
cv_training_data = t.transform_observations(cv_training_data)
cv_test_points = t.transform_observation_features(cv_test_points)
search_space = t.transform_search_space(search_space)
return cv_training_data, cv_test_points, search_space
def _set_kwargs_to_save(
self,
model_key: str,
model_kwargs: Dict[str, Any],
bridge_kwargs: Dict[str, Any],
) -> None:
"""Set properties used to save the model that created a given generator
run, on the `GeneratorRun` object. Each generator run produced by the
`gen` method of this model bridge will have the model key and kwargs
fields set as provided in arguments to this function.
"""
self._model_key = model_key
self._model_kwargs = model_kwargs
self._bridge_kwargs = bridge_kwargs
def _get_serialized_model_state(self) -> Dict[str, Any]:
"""Obtains the state of the underlying model (if using a stateful one)
in a readily JSON-serializable form.
"""
model = not_none(self.model)
return model.serialize_state(raw_state=model._get_state())
def _deserialize_model_state(
self, serialized_state: Dict[str, Any]
) -> Dict[str, Any]:
model = not_none(self.model)
return model.deserialize_state(serialized_state=serialized_state)
[docs] def feature_importances(self, metric_name: str) -> Dict[str, float]:
"""Computes feature importances for a single metric.
Depending on the type of the model, this method will approach sensitivity
analysis (calculating the sensitivity of the metric to changes in the search
space's parameters, a.k.a. features) differently.
For Bayesian optimization models (BoTorch models), this method uses parameter
inverse lengthscales to compute normalized feature importances.
NOTE: Currently, this is only implemented for GP models.
Args:
metric_name: Name of metric to compute feature importances for.
Returns:
A dictionary mapping parameter names to their corresponding feature
importances.
"""
raise NotImplementedError(
"Feature importance not available for this model type"
)
# pyre-fixme[3]: Return annotation cannot be `Any`.
# pyre-fixme[3]: Return annotation cannot be `Any`.
def _transform_observations(self, observations: List[Observation]) -> Any:
"""Apply terminal transform to given observations and return result."""
raise NotImplementedError
# pyre-fixme[3]: Return annotation cannot be `Any`.
# pyre-fixme[3]: Return annotation cannot be `Any`.
def _transform_observation_features(
self, observation_features: List[ObservationFeatures]
) -> Any:
"""Apply terminal transform to given observation features and return result."""
raise NotImplementedError
def __repr__(self) -> str:
return f"{self.__class__.__name__}(model={self.model})"
[docs]def unwrap_observation_data(observation_data: List[ObservationData]) -> TModelPredict:
"""Converts observation data to the format for model prediction outputs.
That format assumes each observation data has the same set of metrics.
"""
metrics = set(observation_data[0].metric_names)
f: TModelMean = {metric: [] for metric in metrics}
cov: TModelCov = {m1: {m2: [] for m2 in metrics} for m1 in metrics}
for od in observation_data:
if set(od.metric_names) != metrics:
raise ValueError(
"Each ObservationData should use same set of metrics. "
"Expected {exp}, got {got}.".format(
exp=metrics, got=set(od.metric_names)
)
)
for i, m1 in enumerate(od.metric_names):
f[m1].append(od.means[i])
for j, m2 in enumerate(od.metric_names):
cov[m1][m2].append(od.covariance[i, j])
return f, cov
[docs]def gen_arms(
observation_features: List[ObservationFeatures],
arms_by_signature: Optional[Dict[str, Arm]] = None,
) -> Tuple[List[Arm], Optional[Dict[str, TCandidateMetadata]]]:
"""Converts observation features to a tuple of arms list and candidate metadata
dict, where arm signatures are mapped to their respective candidate metadata.
"""
# TODO(T34225939): handle static context (which is stored on observation_features)
arms = []
candidate_metadata = {}
for of in observation_features:
arm = Arm(parameters=of.parameters)
if arms_by_signature is not None and arm.signature in arms_by_signature:
existing_arm = arms_by_signature[arm.signature]
arm = Arm(name=existing_arm.name, parameters=existing_arm.parameters)
arms.append(arm)
if of.metadata:
candidate_metadata[arm.signature] = of.metadata
return arms, candidate_metadata or None # None if empty cand. metadata.
[docs]def clamp_observation_features(
observation_features: List[ObservationFeatures], search_space: SearchSpace
) -> List[ObservationFeatures]:
range_parameters = [
p for p in search_space.parameters.values() if isinstance(p, RangeParameter)
]
for obsf in observation_features:
for p in range_parameters:
if p.name not in obsf.parameters:
continue
if p.parameter_type == ParameterType.FLOAT:
val = checked_cast(float, obsf.parameters[p.name])
else:
val = checked_cast(int, obsf.parameters[p.name])
if val < p.lower:
logger.info(
f"Untransformed parameter {val} "
f"less than lower bound {p.lower}, clamping"
)
obsf.parameters[p.name] = p.lower
elif val > p.upper:
logger.info(
f"Untransformed parameter {val} "
f"greater than upper bound {p.upper}, clamping"
)
obsf.parameters[p.name] = p.upper
return observation_features
def _get_status_quo_by_trial(
observations: List[Observation],
status_quo_name: Optional[str] = None,
status_quo_features: Optional[ObservationFeatures] = None,
) -> Optional[Dict[int, ObservationData]]:
r"""
Given a status quo observation, return a dictionary of trial index to
the status quo observation data of each trial.
When either `status_quo_name` or `status_quo_features` exists, return the dict;
when both exist, use `status_quo_name`;
when neither exists, return None.
Args:
observations: List of observations.
status_quo_name: Name of the status quo.
status_quo_features: ObservationFeatures for the status quo.
Returns:
A map from trial index to status quo observation data, or None
"""
trial_idx_to_sq_data = None
if status_quo_name is not None:
# identify status quo by arm name
trial_idx_to_sq_data = {
int(not_none(obs.features.trial_index)): obs.data
for obs in observations
if obs.arm_name == status_quo_name
}
elif status_quo_features is not None:
# identify status quo by (untransformed) feature
status_quo_signature = json.dumps(
status_quo_features.parameters, sort_keys=True
)
trial_idx_to_sq_data = {
int(not_none(obs.features.trial_index)): obs.data
for obs in observations
if json.dumps(obs.features.parameters, sort_keys=True)
== status_quo_signature
}
return trial_idx_to_sq_data
