#!/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
from __future__ import annotations
from collections import OrderedDict
from collections.abc import Iterable, MutableMapping
from datetime import datetime, timedelta
from logging import Logger
from pathlib import Path
from typing import Any, cast
import numpy as np
import pandas as pd
import torch
from ax.core.arm import Arm
from ax.core.auxiliary import AuxiliaryExperiment
from ax.core.base_trial import BaseTrial, TrialStatus
from ax.core.batch_trial import AbandonedArm, BatchTrial
from ax.core.data import Data
from ax.core.experiment import DataType, Experiment
from ax.core.generation_strategy_interface import GenerationStrategyInterface
from ax.core.generator_run import GeneratorRun
from ax.core.map_data import MapData, MapKeyInfo
from ax.core.map_metric import MapMetric
from ax.core.metric import Metric
from ax.core.multi_type_experiment import MultiTypeExperiment
from ax.core.objective import MultiObjective, Objective, ScalarizedObjective
from ax.core.observation import Observation, ObservationData, ObservationFeatures
from ax.core.optimization_config import (
MultiObjectiveOptimizationConfig,
OptimizationConfig,
)
from ax.core.outcome_constraint import (
ObjectiveThreshold,
OutcomeConstraint,
ScalarizedOutcomeConstraint,
)
from ax.core.parameter import (
ChoiceParameter,
FixedParameter,
Parameter,
ParameterType,
RangeParameter,
)
from ax.core.parameter_constraint import (
OrderConstraint,
ParameterConstraint,
SumConstraint,
)
from ax.core.parameter_distribution import ParameterDistribution
from ax.core.risk_measures import RiskMeasure
from ax.core.runner import Runner
from ax.core.search_space import HierarchicalSearchSpace, RobustSearchSpace, SearchSpace
from ax.core.trial import Trial
from ax.core.types import (
ComparisonOp,
TModelCov,
TModelMean,
TModelPredict,
TModelPredictArm,
TParameterization,
TParamValue,
)
from ax.early_stopping.strategies import (
BaseEarlyStoppingStrategy,
PercentileEarlyStoppingStrategy,
ThresholdEarlyStoppingStrategy,
)
from ax.early_stopping.strategies.logical import (
AndEarlyStoppingStrategy,
OrEarlyStoppingStrategy,
)
from ax.exceptions.core import UserInputError
from ax.global_stopping.strategies.base import BaseGlobalStoppingStrategy
from ax.global_stopping.strategies.improvement import ImprovementGlobalStoppingStrategy
from ax.metrics.branin import BraninMetric
from ax.metrics.branin_map import BraninTimestampMapMetric
from ax.metrics.factorial import FactorialMetric
from ax.metrics.hartmann6 import Hartmann6Metric
from ax.modelbridge.factory import Cont_X_trans, get_factorial, get_sobol, Models
from ax.modelbridge.generation_node_input_constructors import (
InputConstructorPurpose,
NodeInputConstructors,
)
from ax.modelbridge.generation_strategy import GenerationNode, GenerationStrategy
from ax.modelbridge.model_spec import ModelSpec
from ax.modelbridge.transition_criterion import (
MaxGenerationParallelism,
MaxTrials,
MinTrials,
TrialBasedCriterion,
)
from ax.models.torch.botorch_modular.acquisition import Acquisition
from ax.models.torch.botorch_modular.kernels import ScaleMaternKernel
from ax.models.torch.botorch_modular.model import BoTorchModel
from ax.models.torch.botorch_modular.sebo import SEBOAcquisition
from ax.models.torch.botorch_modular.surrogate import Surrogate, SurrogateSpec
from ax.models.winsorization_config import WinsorizationConfig
from ax.runners.synthetic import SyntheticRunner
from ax.service.utils.scheduler_options import SchedulerOptions, TrialType
from ax.utils.common.constants import Keys
from ax.utils.common.logger import get_logger
from ax.utils.common.random import set_rng_seed
from ax.utils.common.typeutils import checked_cast
from ax.utils.measurement.synthetic_functions import branin
from botorch.acquisition.acquisition import AcquisitionFunction
from botorch.acquisition.monte_carlo import qExpectedImprovement
from botorch.models.gp_regression import SingleTaskGP
from botorch.models.model import Model
from botorch.models.transforms.input import ChainedInputTransform, Normalize, Round
from botorch.utils.datasets import SupervisedDataset
from botorch.utils.types import DEFAULT
from gpytorch.constraints import Interval
from gpytorch.kernels.rbf_kernel import RBFKernel
from gpytorch.mlls.exact_marginal_log_likelihood import ExactMarginalLogLikelihood
from gpytorch.mlls.marginal_log_likelihood import MarginalLogLikelihood
from gpytorch.priors.torch_priors import GammaPrior, LogNormalPrior
from pyre_extensions import none_throws
logger: Logger = get_logger(__name__)
TEST_SOBOL_SEED = 1234
##############################
# Experiments
##############################
[docs]
def get_experiment(
with_status_quo: bool = True, constrain_search_space: bool = True
) -> Experiment:
return Experiment(
name="test",
search_space=get_search_space(constrain_search_space=constrain_search_space),
optimization_config=get_optimization_config(),
status_quo=get_status_quo() if with_status_quo else None,
description="test description",
tracking_metrics=[Metric(name="tracking")],
is_test=True,
)
[docs]
def get_experiment_with_map_data_type() -> Experiment:
return Experiment(
name="test_map_data",
search_space=get_search_space(),
optimization_config=get_map_optimization_config(),
status_quo=get_status_quo(),
description="test description",
tracking_metrics=[MapMetric(name="tracking")],
is_test=True,
default_data_type=DataType.MAP_DATA,
)
[docs]
def get_trial_based_criterion() -> list[TrialBasedCriterion]:
return [
MaxTrials(
threshold=3,
only_in_statuses=[TrialStatus.RUNNING, TrialStatus.COMPLETED],
not_in_statuses=None,
),
MaxGenerationParallelism(
threshold=5,
only_in_statuses=None,
not_in_statuses=[
TrialStatus.RUNNING,
],
),
]
[docs]
def get_experiment_with_custom_runner_and_metric(
constrain_search_space: bool = True,
immutable: bool = False,
multi_objective: bool = False,
num_trials: int = 3,
has_outcome_constraint: bool = False,
) -> Experiment:
# Create experiment with custom runner and metric
experiment = Experiment(
name="test",
search_space=get_search_space(constrain_search_space=constrain_search_space),
optimization_config=(
get_multi_objective_optimization_config(
custom_metric=True,
outcome_constraint=has_outcome_constraint,
relative=False,
)
if multi_objective
else get_optimization_config(
outcome_constraint=has_outcome_constraint, relative=False
)
),
description="test description",
tracking_metrics=[
CustomTestMetric(name="custom_test_metric", test_attribute="test")
],
runner=CustomTestRunner(test_attribute="test"),
is_test=True,
)
# Create a trial, set its runner and complete it.
for _ in range(num_trials):
sobol_generator = get_sobol(
search_space=experiment.search_space,
)
sobol_run = sobol_generator.gen(
n=1,
optimization_config=(
experiment.optimization_config if not immutable else None
),
)
trial = experiment.new_trial(generator_run=sobol_run)
trial.runner = experiment.runner
trial.mark_running()
data = Data.from_multiple_data(
get_data(
metric_name=metric_name,
trial_index=trial.index,
num_non_sq_arms=len(trial.arms),
include_sq=False,
)
for metric_name in experiment.metrics
)
experiment.attach_data(data)
trial.mark_completed()
if immutable:
experiment._properties = {Keys.IMMUTABLE_SEARCH_SPACE_AND_OPT_CONF: True}
return experiment
[docs]
def get_branin_experiment(
has_optimization_config: bool = True,
with_batch: bool = False,
with_trial: bool = False,
with_status_quo: bool = False,
with_fidelity_parameter: bool = False,
with_choice_parameter: bool = False,
with_str_choice_param: bool = False,
with_parameter_constraint: bool = False,
search_space: SearchSpace | None = None,
minimize: bool = False,
named: bool = True,
num_batch_trial: int = 1,
with_completed_batch: bool = False,
with_completed_trial: bool = False,
) -> Experiment:
search_space = search_space or get_branin_search_space(
with_fidelity_parameter=with_fidelity_parameter,
with_choice_parameter=with_choice_parameter,
with_str_choice_param=with_str_choice_param,
with_parameter_constraint=with_parameter_constraint,
)
exp = Experiment(
name="branin_test_experiment" if named else None,
search_space=search_space,
optimization_config=(
get_branin_optimization_config(minimize=minimize)
if has_optimization_config
else None
),
runner=SyntheticRunner(),
is_test=True,
status_quo=Arm(parameters={"x1": 0.0, "x2": 0.0}) if with_status_quo else None,
)
if with_batch or with_completed_batch:
for _ in range(num_batch_trial):
sobol_generator = get_sobol(search_space=exp.search_space)
sobol_run = sobol_generator.gen(n=15)
trial = exp.new_batch_trial(optimize_for_power=with_status_quo)
trial.add_generator_run(sobol_run)
if with_completed_batch:
trial.mark_running(no_runner_required=True)
exp.attach_data(get_branin_data_batch(batch=trial))
trial.mark_completed()
if with_trial or with_completed_trial:
sobol_generator = get_sobol(search_space=exp.search_space)
sobol_run = sobol_generator.gen(n=1)
trial = exp.new_trial(generator_run=sobol_run)
if with_completed_trial:
trial.mark_running(no_runner_required=True)
exp.attach_data(get_branin_data(trials=[trial])) # Add data for one trial
trial.mark_completed()
return exp
[docs]
def get_branin_experiment_with_status_quo_trials(
num_sobol_trials: int = 5,
multi_objective: bool = False,
) -> tuple[Experiment, ObservationFeatures]:
if multi_objective:
exp = get_branin_experiment_with_multi_objective(
with_batch=True,
with_status_quo=True,
)
else:
exp = get_branin_experiment()
sobol = get_sobol(search_space=exp.search_space)
for _ in range(num_sobol_trials):
sobol_run = sobol.gen(n=1)
t = exp.new_batch_trial().add_generator_run(sobol_run)
t.set_status_quo_with_weight(status_quo=t.arms[0], weight=0.5)
t.run().mark_completed()
status_quo_features = ObservationFeatures(
parameters=exp.trials[0].status_quo.parameters, # pyre-fixme [16]
trial_index=0,
)
return exp, status_quo_features
[docs]
def get_robust_branin_experiment(
risk_measure: RiskMeasure | None = None,
optimization_config: OptimizationConfig | None = None,
num_sobol_trials: int = 2,
) -> Experiment:
x1_dist = ParameterDistribution(
parameters=["x1"], distribution_class="norm", distribution_parameters={}
)
search_space = RobustSearchSpace(
parameters=[
RangeParameter(
name="x1", parameter_type=ParameterType.FLOAT, lower=-5, upper=10
),
RangeParameter(
name="x2", parameter_type=ParameterType.FLOAT, lower=0, upper=15
),
],
parameter_distributions=[x1_dist],
num_samples=16,
)
risk_measure = risk_measure or RiskMeasure(
risk_measure="CVaR",
options={"n_w": 16, "alpha": 0.8},
)
optimization_config = optimization_config or OptimizationConfig(
objective=Objective(
metric=BraninMetric(
name="branin_metric", param_names=["x1", "x2"], lower_is_better=True
),
minimize=True,
),
risk_measure=risk_measure,
)
exp = Experiment(
name="branin_experiment",
search_space=search_space,
optimization_config=optimization_config,
runner=SyntheticRunner(),
)
sobol = get_sobol(search_space=exp.search_space)
for _ in range(num_sobol_trials):
exp.new_trial(generator_run=sobol.gen(1)).run().mark_completed()
return exp
[docs]
def get_map_metric(name: str, rate: float | None = None) -> BraninTimestampMapMetric:
return BraninTimestampMapMetric(
name=name,
param_names=["x1", "x2"],
rate=rate,
lower_is_better=True,
)
[docs]
def get_branin_experiment_with_timestamp_map_metric(
with_status_quo: bool = False,
rate: float | None = None,
map_tracking_metric: bool = False,
) -> Experiment:
tracking_metric = (
get_map_metric(name="tracking_branin_map", rate=rate)
if map_tracking_metric
else BraninMetric(name="branin", param_names=["x1", "x2"], lower_is_better=True)
)
exp = Experiment(
name="branin_with_timestamp_map_metric",
search_space=get_branin_search_space(),
optimization_config=OptimizationConfig(
objective=Objective(
metric=get_map_metric(name="branin_map", rate=rate), minimize=True
)
),
tracking_metrics=[tracking_metric],
runner=SyntheticRunner(),
default_data_type=DataType.MAP_DATA,
)
if with_status_quo:
exp.status_quo = Arm(parameters={"x1": 0.0, "x2": 0.0})
return exp
[docs]
def run_branin_experiment_with_generation_strategy(
generation_strategy: GenerationStrategy,
num_trials: int = 6,
kwargs_for_get_branin_experiment: dict[str, Any] | None = None,
) -> Experiment:
"""Gets a Branin experiment using any given kwargs and runs
num_trials trials using the given generation strategy."""
kwargs_for_get_branin_experiment = kwargs_for_get_branin_experiment or {}
exp = get_branin_experiment(**kwargs_for_get_branin_experiment)
for _ in range(num_trials):
gr = generation_strategy.gen(n=1, experiment=exp)
trial = exp.new_trial(generator_run=gr)
trial.mark_running(no_runner_required=True)
exp.attach_data(get_branin_data(trials=[trial]))
trial.mark_completed()
return exp
[docs]
def get_test_map_data_experiment(
num_trials: int,
num_fetches: int,
num_complete: int,
map_tracking_metric: bool = False,
) -> Experiment:
experiment = get_branin_experiment_with_timestamp_map_metric(
rate=0.5, map_tracking_metric=map_tracking_metric
)
for i in range(num_trials):
trial = experiment.new_trial().add_arm(arm=get_branin_arms(n=1, seed=i)[0])
trial.run()
for _ in range(num_fetches):
# each time we call fetch, we grab another timestamp
experiment.fetch_data()
for i in range(num_complete):
experiment.trials[i].mark_as(status=TrialStatus.COMPLETED)
return experiment
[docs]
def get_multi_type_experiment(
add_trial_type: bool = True, add_trials: bool = False, num_arms: int = 10
) -> MultiTypeExperiment:
oc = OptimizationConfig(Objective(BraninMetric("m1", ["x1", "x2"]), minimize=True))
experiment = MultiTypeExperiment(
name="test_exp",
search_space=get_branin_search_space(),
default_trial_type="type1",
default_runner=SyntheticRunner(dummy_metadata="dummy1"),
optimization_config=oc,
)
experiment.add_trial_type(
trial_type="type2", runner=SyntheticRunner(dummy_metadata="dummy2")
)
# Switch the order of variables so metric gives different results
experiment.add_tracking_metric(
BraninMetric("m2", ["x2", "x1"]), trial_type="type2", canonical_name="m1"
)
if add_trials and add_trial_type:
generator = get_sobol(experiment.search_space)
gr = generator.gen(num_arms)
t1 = experiment.new_batch_trial(generator_run=gr, trial_type="type1")
t2 = experiment.new_batch_trial(generator_run=gr, trial_type="type2")
t1.set_status_quo_with_weight(status_quo=t1.arms[0], weight=0.5)
t2.set_status_quo_with_weight(status_quo=t2.arms[0], weight=0.5)
t1.run()
t2.run()
return experiment
[docs]
def get_multi_type_experiment_with_multi_objective(
add_trials: bool = False,
) -> MultiTypeExperiment:
oc = get_branin_multi_objective_optimization_config()
experiment = MultiTypeExperiment(
name="test_exp",
search_space=get_branin_search_space(),
default_trial_type="type1",
default_runner=SyntheticRunner(dummy_metadata="dummy1"),
optimization_config=oc,
)
experiment.add_trial_type(
trial_type="type2", runner=SyntheticRunner(dummy_metadata="dummy2")
)
if add_trials:
generator = get_sobol(experiment.search_space)
gr = generator.gen(10)
t1 = experiment.new_batch_trial(generator_run=gr, trial_type="type1")
t2 = experiment.new_batch_trial(generator_run=gr, trial_type="type2")
t1.set_status_quo_with_weight(status_quo=t1.arms[0], weight=0.5)
t2.set_status_quo_with_weight(status_quo=t2.arms[0], weight=0.5)
t1.run()
t2.run()
return experiment
[docs]
def get_factorial_experiment(
has_optimization_config: bool = True,
with_batch: bool = False,
with_status_quo: bool = False,
) -> Experiment:
exp = Experiment(
name="factorial_test_experiment",
search_space=get_factorial_search_space(),
optimization_config=(
OptimizationConfig(
objective=Objective(metric=get_factorial_metric(), minimize=False)
)
if has_optimization_config
else None
),
runner=SyntheticRunner(),
is_test=True,
tracking_metrics=[get_factorial_metric("secondary_metric")],
)
if with_status_quo:
exp.status_quo = Arm(
parameters={
"factor1": "level11",
"factor2": "level21",
"factor3": "level31",
}
)
if with_batch:
factorial_generator = get_factorial(search_space=exp.search_space)
factorial_run = factorial_generator.gen(n=-1)
exp.new_batch_trial(optimize_for_power=with_status_quo).add_generator_run(
factorial_run
)
return exp
[docs]
def get_experiment_with_repeated_arms(num_repeated_arms: int) -> Experiment:
batch_trial = get_batch_trial_with_repeated_arms(num_repeated_arms)
return batch_trial.experiment
[docs]
def get_experiment_with_trial() -> Experiment:
trial = get_trial()
return trial.experiment
[docs]
def get_experiment_with_batch_trial(constrain_search_space: bool = True) -> Experiment:
batch_trial = get_batch_trial(constrain_search_space=constrain_search_space)
return batch_trial.experiment
[docs]
def get_experiment_with_batch_and_single_trial() -> Experiment:
batch_trial = get_batch_trial()
batch_trial.experiment.new_trial(generator_run=GeneratorRun(arms=[get_arm()]))
return batch_trial.experiment
[docs]
def get_experiment_with_trial_with_ttl() -> Experiment:
batch_trial = get_batch_trial()
batch_trial.experiment.new_trial(
generator_run=GeneratorRun(arms=[get_arm()]), ttl_seconds=1
)
return batch_trial.experiment
[docs]
def get_experiment_with_data() -> Experiment:
batch_trial = get_batch_trial()
batch_trial.experiment.attach_data(data=get_data())
batch_trial.experiment.attach_data(data=get_data())
batch_trial.experiment.attach_data(data=get_data())
return batch_trial.experiment
[docs]
def get_experiment_with_map_data() -> Experiment:
experiment = get_experiment_with_map_data_type()
experiment.new_trial()
experiment.add_tracking_metric(MapMetric("ax_test_metric"))
experiment.attach_data(data=get_map_data())
return experiment
[docs]
def get_experiment_with_multi_objective() -> Experiment:
optimization_config = get_multi_objective_optimization_config()
exp = Experiment(
name="test_experiment_multi_objective",
search_space=get_branin_search_space(),
optimization_config=optimization_config,
description="test experiment with multi objective",
runner=SyntheticRunner(),
tracking_metrics=[Metric(name="tracking")],
is_test=True,
)
return exp
[docs]
def get_branin_experiment_with_multi_objective(
has_optimization_config: bool = True,
has_objective_thresholds: bool = False,
with_batch: bool = False,
with_status_quo: bool = False,
with_fidelity_parameter: bool = False,
num_objectives: int = 2,
with_trial: bool = False,
with_completed_trial: bool = False,
with_relative_constraint: bool = False,
with_absolute_constraint: bool = False,
) -> Experiment:
exp = Experiment(
name="branin_test_experiment",
search_space=get_branin_search_space(
with_fidelity_parameter=with_fidelity_parameter
),
optimization_config=(
get_branin_multi_objective_optimization_config(
has_objective_thresholds=has_objective_thresholds,
num_objectives=num_objectives,
with_relative_constraint=with_relative_constraint,
with_absolute_constraint=with_absolute_constraint,
)
if has_optimization_config
else None
),
runner=SyntheticRunner(),
is_test=True,
)
if with_status_quo:
# Experiment chooses the name "status_quo" by default
exp.status_quo = Arm(parameters={"x1": 0.0, "x2": 0.0})
if with_batch:
sobol_generator = get_sobol(search_space=exp.search_space, seed=TEST_SOBOL_SEED)
sobol_run = sobol_generator.gen(n=5)
exp.new_batch_trial(optimize_for_power=with_status_quo).add_generator_run(
sobol_run
)
if with_trial or with_completed_trial:
sobol_generator = get_sobol(search_space=exp.search_space)
sobol_run = sobol_generator.gen(n=1)
trial = exp.new_trial(generator_run=sobol_run)
if with_completed_trial:
trial.mark_running(no_runner_required=True)
exp.attach_data(
get_branin_data_multi_objective(trial_indices=[trial.index])
) # Add data for one trial
trial.mark_completed()
return exp
[docs]
def get_branin_with_multi_task(with_multi_objective: bool = False) -> Experiment:
exp = Experiment(
name="branin_test_experiment",
search_space=get_branin_search_space(),
optimization_config=(
get_branin_multi_objective_optimization_config(
has_objective_thresholds=True,
)
if with_multi_objective
else get_branin_optimization_config()
),
runner=SyntheticRunner(),
is_test=True,
)
exp.status_quo = Arm(parameters={"x1": 0.0, "x2": 0.0}, name="status_quo")
sobol_generator = get_sobol(search_space=exp.search_space, seed=TEST_SOBOL_SEED)
sobol_run = sobol_generator.gen(n=5)
exp.new_batch_trial(optimize_for_power=True).add_generator_run(sobol_run)
none_throws(exp.trials.get(0)).run()
exp.new_batch_trial(optimize_for_power=True).add_generator_run(sobol_run)
none_throws(exp.trials.get(1)).run()
return exp
[docs]
def get_experiment_with_scalarized_objective_and_outcome_constraint() -> Experiment:
objective = get_scalarized_objective()
outcome_constraints = [
get_outcome_constraint(),
get_scalarized_outcome_constraint(),
]
optimization_config = OptimizationConfig(
objective=objective, outcome_constraints=outcome_constraints
)
return Experiment(
name="test_experiment_scalarized_objective and outcome constraint",
search_space=get_search_space(),
optimization_config=optimization_config,
status_quo=get_status_quo(),
description="test experiment with scalarized objective and outcome constraint",
tracking_metrics=[Metric(name="tracking")],
is_test=True,
)
[docs]
def get_hierarchical_search_space_experiment(
num_observations: int = 0,
) -> Experiment:
experiment = Experiment(
name="test_experiment_hss",
description="test experiment with hierarchical search space",
search_space=get_hierarchical_search_space(),
optimization_config=get_optimization_config(),
)
sobol_generator = get_sobol(search_space=experiment.search_space)
for i in range(num_observations):
trial = experiment.new_trial(generator_run=sobol_generator.gen(1))
trial.mark_running(no_runner_required=True)
data = Data(
df=pd.DataFrame.from_records(
[
{
"arm_name": f"{i}_0",
"metric_name": f"m{j + 1}",
"mean": o,
"sem": None,
"trial_index": i,
}
for j, o in enumerate(torch.rand(2).tolist())
]
)
)
experiment.attach_data(data)
trial.mark_completed()
return experiment
[docs]
def get_experiment_with_observations(
observations: list[list[float]],
minimize: bool = False,
scalarized: bool = False,
constrained: bool = False,
with_tracking_metrics: bool = False,
search_space: SearchSpace | None = None,
) -> Experiment:
if observations:
multi_objective = (len(observations[0]) - constrained) > 1
else:
multi_objective = False
if multi_objective:
metrics = [
Metric(name="m1", lower_is_better=minimize),
Metric(name="m2", lower_is_better=False),
]
if scalarized:
optimization_config = OptimizationConfig(
objective=ScalarizedObjective(metrics)
)
if constrained:
raise NotImplementedError
else:
optimization_config = MultiObjectiveOptimizationConfig(
objective=MultiObjective(metrics=metrics),
objective_thresholds=[
ObjectiveThreshold(
metric=metrics[i],
bound=0.0,
relative=False,
op=(
ComparisonOp.LEQ
if metrics[i].lower_is_better
else ComparisonOp.GEQ
),
)
for i in [0, 1]
],
outcome_constraints=(
[
OutcomeConstraint(
metric=Metric(name="m3"),
op=ComparisonOp.GEQ,
bound=0.0,
relative=False,
)
]
if constrained
else None
),
)
else:
if scalarized:
raise NotImplementedError
objective = Objective(metric=Metric(name="m1"), minimize=minimize)
if constrained:
constraint = OutcomeConstraint(
metric=Metric(name="m2"),
op=ComparisonOp.GEQ,
bound=0.0,
relative=False,
)
optimization_config = OptimizationConfig(
objective=objective, outcome_constraints=[constraint]
)
else:
optimization_config = OptimizationConfig(objective=objective)
search_space = search_space or get_search_space_for_range_values(min=0.0, max=1.0)
exp = Experiment(
search_space=search_space,
optimization_config=optimization_config,
tracking_metrics=(
[Metric(name=f"m{len(observations[0])}", lower_is_better=False)]
if with_tracking_metrics
else None
),
runner=SyntheticRunner(),
is_test=True,
)
sobol_generator = get_sobol(search_space=search_space)
for i, obs in enumerate(observations):
# Create a dummy trial to add the observation.
trial = exp.new_trial(generator_run=sobol_generator.gen(n=1))
data = Data(
df=pd.DataFrame.from_records(
[
{
"arm_name": f"{i}_0",
"metric_name": f"m{j + 1}",
"mean": o,
"sem": None,
"trial_index": i,
}
for j, o in enumerate(obs)
]
)
)
exp.attach_data(data)
trial.run().complete()
return exp
[docs]
def get_high_dimensional_branin_experiment(
with_batch: bool = False, with_status_quo: bool = False
) -> Experiment:
search_space = SearchSpace(
# pyre-fixme[6]: In call `SearchSpace.__init__`, for 1st parameter `parameters`
# expected `List[Parameter]` but got `List[RangeParameter]`.
parameters=[
RangeParameter(
name=f"x{i}",
parameter_type=ParameterType.FLOAT,
lower=-5.0,
upper=10.0,
)
for i in range(25)
]
+ [
RangeParameter(
name=f"x{i + 25}",
parameter_type=ParameterType.FLOAT,
lower=0.0,
upper=15.0,
)
for i in range(25)
],
)
sq_parameters = {f"x{i}": 1.0 if i < 25 else 2.0 for i in range(50)}
optimization_config = OptimizationConfig(
objective=Objective(
metric=BraninMetric(
name="objective",
param_names=["x19", "x44"],
),
minimize=True,
)
)
exp = Experiment(
name="high_dimensional_branin_experiment",
search_space=search_space,
optimization_config=optimization_config,
runner=SyntheticRunner(),
# pyre-fixme[6]: expects a union type for val instead of defined {str: float}
status_quo=Arm(sq_parameters) if with_status_quo else None,
)
if with_batch:
sobol_generator = get_sobol(search_space=exp.search_space)
sobol_run = sobol_generator.gen(n=15)
exp.new_batch_trial().add_generator_run(sobol_run)
return exp
[docs]
def get_auxiliary_experiment() -> AuxiliaryExperiment:
return AuxiliaryExperiment(experiment=get_experiment_with_data())
##############################
# Search Spaces
##############################
[docs]
def get_search_space(constrain_search_space: bool = True) -> SearchSpace:
parameters: list[Parameter] = [
get_range_parameter(),
get_range_parameter2(),
get_choice_parameter(),
get_fixed_parameter(),
]
parameter_constraints = []
if constrain_search_space:
parameter_constraints = [
get_order_constraint(),
get_parameter_constraint(),
get_sum_constraint1(),
]
return SearchSpace(
parameters=parameters,
parameter_constraints=parameter_constraints,
)
[docs]
def get_branin_search_space(
with_fidelity_parameter: bool = False,
with_choice_parameter: bool = False,
with_str_choice_param: bool = False,
with_parameter_constraint: bool = False,
) -> SearchSpace:
parameters = [
RangeParameter(
name="x1", parameter_type=ParameterType.FLOAT, lower=-5, upper=10
),
(
ChoiceParameter(
name="x2",
parameter_type=ParameterType.FLOAT,
values=[float(x) for x in range(0, 16)],
)
if with_choice_parameter
else RangeParameter(
name="x2", parameter_type=ParameterType.FLOAT, lower=0, upper=15
)
),
]
if with_str_choice_param:
parameters.append(
ChoiceParameter(
name="str_param",
parameter_type=ParameterType.STRING,
values=["foo", "bar", "baz"],
)
)
if with_fidelity_parameter:
parameters.append(
RangeParameter(
name="fidelity",
parameter_type=ParameterType.FLOAT,
lower=0.0,
upper=1.0,
is_fidelity=True,
target_value=1.0,
)
)
if with_parameter_constraint:
constraints = [
ParameterConstraint(constraint_dict={"x1": 1, "x2": 1}, bound=15.0)
]
else:
constraints = None
return SearchSpace(
parameters=cast(list[Parameter], parameters), parameter_constraints=constraints
)
[docs]
def get_factorial_search_space() -> SearchSpace:
return SearchSpace(
parameters=[
ChoiceParameter(
name="factor1",
parameter_type=ParameterType.STRING,
values=["level11", "level12", "level13"],
),
ChoiceParameter(
name="factor2",
parameter_type=ParameterType.STRING,
values=["level21", "level22"],
),
ChoiceParameter(
name="factor3",
parameter_type=ParameterType.STRING,
values=["level31", "level32", "level33", "level34"],
),
]
)
[docs]
def get_large_factorial_search_space(
num_levels: int = 10, num_parameters: int = 6
) -> SearchSpace:
return SearchSpace(
parameters=[
ChoiceParameter(
name=f"factor{j}",
parameter_type=ParameterType.STRING,
values=[f"level1{i}" for i in range(num_levels)],
)
for j in range(num_parameters)
]
)
[docs]
def get_large_ordinal_search_space(
n_ordinal_choice_parameters: int,
n_continuous_range_parameters: int,
) -> SearchSpace:
return SearchSpace(
parameters=[ # pyre-ignore[6]
RangeParameter(
name=f"x{i}",
parameter_type=ParameterType.FLOAT,
lower=0.0,
upper=1.0,
)
for i in range(n_continuous_range_parameters)
]
+ [
ChoiceParameter(
name=f"y{i}",
parameter_type=ParameterType.INT,
values=[2, 4, 8, 16],
)
for i in range(n_ordinal_choice_parameters)
]
)
[docs]
def get_hartmann_search_space(with_fidelity_parameter: bool = False) -> SearchSpace:
parameters = [
RangeParameter(
name=f"x{idx + 1}", parameter_type=ParameterType.FLOAT, lower=0.0, upper=1.0
)
for idx in range(6)
]
if with_fidelity_parameter:
parameters.append(
RangeParameter(
name="fidelity",
parameter_type=ParameterType.FLOAT,
lower=0.0,
upper=1.0,
is_fidelity=True,
target_value=1.0,
)
)
return SearchSpace(parameters=cast(list[Parameter], parameters))
[docs]
def get_search_space_for_value(val: float = 3.0) -> SearchSpace:
return SearchSpace([FixedParameter("x", ParameterType.FLOAT, val)])
[docs]
def get_search_space_for_range_value(min: float = 3.0, max: float = 6.0) -> SearchSpace:
return SearchSpace([RangeParameter("x", ParameterType.FLOAT, min, max)])
[docs]
def get_search_space_for_range_values(
min: float = 3.0, max: float = 6.0
) -> SearchSpace:
return SearchSpace(
[
RangeParameter("x", ParameterType.FLOAT, min, max),
RangeParameter("y", ParameterType.FLOAT, min, max),
]
)
[docs]
def get_discrete_search_space() -> SearchSpace:
return SearchSpace(
[
RangeParameter("x", ParameterType.INT, 0, 3),
RangeParameter("y", ParameterType.INT, 5, 7),
ChoiceParameter("z", ParameterType.STRING, ["red", "panda", "bear"]),
]
)
[docs]
def get_small_discrete_search_space() -> SearchSpace:
return SearchSpace(
[
RangeParameter("x", ParameterType.INT, 0, 1),
ChoiceParameter("y", ParameterType.STRING, ["red", "panda"]),
]
)
[docs]
def get_search_space_with_choice_parameters(
num_ordered_parameters: int = 2,
num_unordered_choices: int = 5,
) -> SearchSpace:
parameters = []
for i in range(num_ordered_parameters):
parameters.append(
ChoiceParameter(
name=f"ordered_{i}",
parameter_type=ParameterType.INT,
values=list(range(10)), # pyre-ignore
is_ordered=True,
)
)
parameters.append(
ChoiceParameter(
name="unordered",
parameter_type=ParameterType.INT,
values=list(range(num_unordered_choices)), # pyre-ignore
is_ordered=False,
)
)
return SearchSpace(parameters=parameters)
[docs]
def get_hierarchical_search_space(
with_fixed_parameter: bool = False,
) -> HierarchicalSearchSpace:
parameters: list[Parameter] = [
get_model_parameter(with_fixed_parameter=with_fixed_parameter),
get_lr_parameter(),
get_l2_reg_weight_parameter(),
get_num_boost_rounds_parameter(),
]
if with_fixed_parameter:
parameters.append(get_fixed_parameter())
return HierarchicalSearchSpace(parameters=parameters)
[docs]
def get_robust_search_space(
lb: float = 0.0,
ub: float = 5.0,
multivariate: bool = False,
use_discrete: bool = False,
num_samples: int = 4, # dummy
) -> RobustSearchSpace:
parameters = [
RangeParameter("x", ParameterType.FLOAT, lb, ub),
RangeParameter("y", ParameterType.FLOAT, lb, ub),
RangeParameter("z", ParameterType.INT, lb, ub),
ChoiceParameter("c", ParameterType.STRING, ["red", "blue", "green"]),
]
if multivariate:
if use_discrete:
raise UserInputError(
"`multivariate` and `use_discrete` are not supported together."
)
distributions = [
ParameterDistribution(
parameters=["x", "y"],
distribution_class="multivariate_normal",
distribution_parameters={
"mean": [0.1, 0.2],
"cov": [[0.5, 0.1], [0.1, 0.3]],
},
)
]
else:
distributions = []
distributions.append(
ParameterDistribution(
parameters=["x"],
distribution_class="norm",
distribution_parameters={"loc": 1.0, "scale": ub - lb},
)
)
if use_discrete:
distributions.append(
ParameterDistribution(
parameters=["z"],
distribution_class="binom",
distribution_parameters={"n": 20, "p": 0.5},
)
)
else:
distributions.append(
ParameterDistribution(
parameters=["y"],
distribution_class="expon",
distribution_parameters={},
)
)
return RobustSearchSpace(
# pyre-ignore Incompatible parameter type [6]
parameters=parameters,
parameter_distributions=distributions,
num_samples=num_samples,
)
[docs]
def get_robust_search_space_environmental(
lb: float = 0.0,
ub: float = 5.0,
) -> RobustSearchSpace:
parameters = [
RangeParameter("x", ParameterType.FLOAT, lb, ub),
RangeParameter("y", ParameterType.FLOAT, lb, ub),
]
environmental_variables = [
RangeParameter("z", ParameterType.INT, lb, ub),
]
distributions = [
ParameterDistribution(
parameters=["z"],
distribution_class="binom",
distribution_parameters={"n": 5, "p": 0.5},
)
]
return RobustSearchSpace(
# pyre-ignore Incompatible parameter type [6]
parameters=parameters,
parameter_distributions=distributions,
num_samples=4,
# pyre-ignore Incompatible parameter type [6]
environmental_variables=environmental_variables,
)
##############################
# Trials
##############################
[docs]
def get_batch_trial(
abandon_arm: bool = True,
experiment: Experiment | None = None,
constrain_search_space: bool = True,
) -> BatchTrial:
experiment = experiment or get_experiment(
constrain_search_space=constrain_search_space
)
batch = experiment.new_batch_trial()
arms = get_arms_from_dict(get_arm_weights1())
weights = get_weights_from_dict(get_arm_weights1())
batch.add_arms_and_weights(arms=arms, weights=weights, multiplier=0.75)
if abandon_arm:
batch.mark_arm_abandoned(batch.arms[0].name, "abandoned reason")
batch.runner = SyntheticRunner()
batch.set_status_quo_with_weight(status_quo=arms[0], weight=0.5)
batch._generation_step_index = 0
return batch
[docs]
def get_batch_trial_with_repeated_arms(num_repeated_arms: int) -> BatchTrial:
"""Create a batch that contains both new arms and N arms from the last
existed trial in the experiment. Where N is equal to the input argument
'num_repeated_arms'.
"""
experiment = get_experiment_with_batch_trial()
if len(experiment.trials) > 0:
# Get last (previous) trial.
prev_trial = experiment.trials[len(experiment.trials) - 1]
# Take the first N arms, where N is num_repeated_arms.
if len(prev_trial.arms) < num_repeated_arms:
logger.warning(
"There are less arms in the previous trial than the value of "
"input parameter 'num_repeated_arms'. Thus all the arms from "
"the last trial will be repeated in the new trial."
)
prev_arms = prev_trial.arms[:num_repeated_arms]
if isinstance(prev_trial, BatchTrial):
prev_weights = prev_trial.weights[:num_repeated_arms]
else:
prev_weights = [1] * len(prev_arms)
else:
raise Exception(
"There are no previous trials in this experiment. Thus the new "
"batch was not created as no repeated arms could be added."
)
# Create new (next) arms.
next_arms = get_arms_from_dict(get_arm_weights2())
next_weights = get_weights_from_dict(get_arm_weights2())
# Add num_repeated_arms to the new trial.
arms = prev_arms + next_arms
weights = prev_weights + next_weights
batch = experiment.new_batch_trial()
batch.add_arms_and_weights(arms=arms, weights=weights, multiplier=1)
batch.runner = SyntheticRunner()
batch.set_status_quo_with_weight(status_quo=arms[0], weight=0.5)
return batch
[docs]
def get_trial() -> Trial:
experiment = get_experiment()
trial = experiment.new_trial(ttl_seconds=72)
arm = get_arms_from_dict(get_arm_weights1())[0]
trial.add_arm(arm)
trial.runner = SyntheticRunner()
trial._generation_step_index = 0
trial.update_run_metadata({"workflow_run_id": [12345]})
return trial
[docs]
def get_hss_trials_with_fixed_parameter(exp: Experiment) -> dict[int, BaseTrial]:
return {
0: Trial(experiment=exp).add_arm(
arm=Arm(
parameters={
"model": "Linear",
"learning_rate": 0.05,
"l2_reg_weight": 1e-4,
"num_boost_rounds": 15,
"z": True,
},
name="0_0",
)
),
1: Trial(experiment=exp).add_arm(
arm=Arm(
parameters={
"model": "XGBoost",
"learning_rate": 0.05,
"l2_reg_weight": 1e-4,
"num_boost_rounds": 15,
"z": True,
},
name="1_0",
)
),
}
[docs]
class TestTrial(BaseTrial):
"Trial class to test unsupported trial type error"
_arms: list[Arm] = []
def __repr__(self) -> str:
return "test"
def _get_candidate_metadata(self, arm_name: str) -> dict[str, Any] | None:
return None
def _get_candidate_metadata_from_all_generator_runs(
self,
) -> dict[str, dict[str, Any] | None]:
return {"test": None}
[docs]
def abandoned_arms(self) -> str:
return "test"
@property
def arms(self) -> list[Arm]:
return self._arms
@arms.setter
def arms(self, val: list[Arm]) -> None:
self._arms = val
[docs]
def arms_by_name(self) -> str:
return "test"
[docs]
def generator_runs(self) -> str:
return "test"
##############################
# Parameters
##############################
[docs]
def get_range_parameter() -> RangeParameter:
return RangeParameter(
name="w",
parameter_type=ParameterType.FLOAT,
lower=0.5,
upper=5.5,
log_scale=False,
digits=5,
)
[docs]
def get_range_parameter2() -> RangeParameter:
return RangeParameter(name="x", parameter_type=ParameterType.INT, lower=1, upper=10)
[docs]
def get_choice_parameter() -> ChoiceParameter:
return ChoiceParameter(
name="y",
parameter_type=ParameterType.STRING,
# Expected `List[typing.Optional[typing.Union[bool, float, str]]]` for 4th
# parameter `values` to call
# `ax.core.parameter.ChoiceParameter.__init__` but got `List[str]`.
values=["foo", "bar", "baz"],
)
[docs]
def get_ordered_choice_parameter() -> ChoiceParameter:
return ChoiceParameter(
name="y",
parameter_type=ParameterType.INT,
# Expected `List[typing.Optional[typing.Union[bool, float, str]]]` for 4th
# parameter `values` to call
# `ax.core.parameter.ChoiceParameter.__init__` but got `List[str]`.
values=[1, 2, 3],
is_ordered=True,
)
[docs]
def get_task_choice_parameter() -> ChoiceParameter:
return ChoiceParameter(
name="y",
parameter_type=ParameterType.INT,
values=[1, 2, 3],
is_task=True,
target_value=1,
)
[docs]
def get_fixed_parameter() -> FixedParameter:
return FixedParameter(name="z", parameter_type=ParameterType.BOOL, value=True)
[docs]
def get_model_parameter(with_fixed_parameter: bool = False) -> ChoiceParameter:
return ChoiceParameter(
name="model",
parameter_type=ParameterType.STRING,
values=["Linear", "XGBoost"],
dependents={
"Linear": ["learning_rate", "l2_reg_weight"],
"XGBoost": (
["num_boost_rounds", "z"]
if with_fixed_parameter
else ["num_boost_rounds"]
),
},
)
[docs]
def get_lr_parameter() -> RangeParameter:
return RangeParameter(
name="learning_rate",
parameter_type=ParameterType.FLOAT,
lower=0.001,
upper=0.1,
)
[docs]
def get_l2_reg_weight_parameter() -> RangeParameter:
return RangeParameter(
name="l2_reg_weight",
parameter_type=ParameterType.FLOAT,
lower=0.00001,
upper=0.001,
)
[docs]
def get_num_boost_rounds_parameter() -> RangeParameter:
return RangeParameter(
name="num_boost_rounds",
parameter_type=ParameterType.INT,
lower=10,
upper=20,
)
##############################
# Parameter Constraints
##############################
[docs]
def get_order_constraint() -> OrderConstraint:
w = get_range_parameter()
x = get_range_parameter2()
return OrderConstraint(lower_parameter=x, upper_parameter=w)
[docs]
def get_parameter_constraint(
param_x: str = "x", param_y: str = "w"
) -> ParameterConstraint:
return ParameterConstraint(constraint_dict={param_x: 1.0, param_y: -1.0}, bound=1.0)
[docs]
def get_sum_constraint1() -> SumConstraint:
w = get_range_parameter()
x = get_range_parameter2()
return SumConstraint(parameters=[x, w], is_upper_bound=False, bound=10.0)
[docs]
def get_sum_constraint2() -> SumConstraint:
w = get_range_parameter()
x = get_range_parameter2()
return SumConstraint(parameters=[x, w], is_upper_bound=True, bound=10.0)
##############################
# Metrics
##############################
[docs]
def get_metric() -> Metric:
return Metric(name="m1", properties={"prop": "val"})
[docs]
def get_branin_metric(
name: str = "branin", lower_is_better: bool = True
) -> BraninMetric:
param_names = ["x1", "x2"]
return BraninMetric(
name=name,
param_names=param_names,
noise_sd=0.01,
lower_is_better=lower_is_better,
)
[docs]
def get_hartmann_metric(name: str = "hartmann") -> Hartmann6Metric:
param_names = [f"x{idx + 1}" for idx in range(6)]
return Hartmann6Metric(name=name, param_names=param_names, noise_sd=0.01)
[docs]
def get_factorial_metric(name: str = "success_metric") -> FactorialMetric:
coefficients: dict[str, dict[TParamValue, float]] = {
"factor1": {"level11": 0.1, "level12": 0.2, "level13": 0.3},
"factor2": {"level21": 0.1, "level22": 0.2},
"factor3": {"level31": 0.1, "level32": 0.2, "level33": 0.3, "level34": 0.4},
}
return FactorialMetric(name=name, coefficients=coefficients, batch_size=int(1e4))
##############################
# Outcome Constraints
##############################
[docs]
def get_objective_threshold(
metric_name: str = "m1",
bound: float = -0.25,
comparison_op: ComparisonOp = ComparisonOp.GEQ,
) -> ObjectiveThreshold:
return ObjectiveThreshold(
metric=Metric(name=metric_name), bound=bound, op=comparison_op
)
[docs]
def get_outcome_constraint(
metric: Metric | None = None, relative: bool = True, bound: float = -0.25
) -> OutcomeConstraint:
if metric is None:
metric = Metric(name="m2")
return OutcomeConstraint(
metric=metric, op=ComparisonOp.GEQ, bound=bound, relative=relative
)
[docs]
def get_scalarized_outcome_constraint() -> ScalarizedOutcomeConstraint:
return ScalarizedOutcomeConstraint(
metrics=[Metric(name="oc_m3"), Metric(name="oc_m4")],
weights=[0.2, 0.8],
op=ComparisonOp.GEQ,
bound=-0.25,
)
[docs]
def get_branin_outcome_constraint(name: str = "branin") -> OutcomeConstraint:
return OutcomeConstraint(
metric=get_branin_metric(name=name), op=ComparisonOp.LEQ, bound=0.0
)
##############################
# Objectives
##############################
[docs]
def get_objective(minimize: bool = False) -> Objective:
return Objective(metric=Metric(name="m1"), minimize=minimize)
[docs]
def get_map_objective(minimize: bool = False) -> Objective:
return Objective(metric=MapMetric(name="m1"), minimize=minimize)
[docs]
def get_multi_objective() -> MultiObjective:
return MultiObjective(
objectives=[
Objective(metric=Metric(name="m1"), minimize=False),
Objective(metric=Metric(name="m3", lower_is_better=True), minimize=True),
],
)
[docs]
def get_custom_multi_objective() -> MultiObjective:
return MultiObjective(
objectives=[
Objective(
metric=CustomTestMetric(name="m1", test_attribute="test"),
minimize=False,
),
Objective(
metric=CustomTestMetric(
name="m3", lower_is_better=True, test_attribute="test"
),
minimize=True,
),
],
)
[docs]
def get_many_branin_objective_opt_config(
n_objectives: int,
) -> MultiObjectiveOptimizationConfig:
return MultiObjectiveOptimizationConfig(
objective=MultiObjective(
objectives=[get_branin_objective(name=f"m{i}") for i in range(n_objectives)]
)
)
[docs]
def get_scalarized_objective() -> Objective:
return ScalarizedObjective(
metrics=[Metric(name="m1"), Metric(name="m3")],
weights=[1.0, 2.0],
minimize=False,
)
[docs]
def get_branin_objective(name: str = "branin", minimize: bool = False) -> Objective:
return Objective(
metric=get_branin_metric(name=name, lower_is_better=minimize), minimize=minimize
)
[docs]
def get_branin_multi_objective(num_objectives: int = 2) -> MultiObjective:
_validate_num_objectives(num_objectives=num_objectives)
objectives = [
Objective(metric=get_branin_metric(name="branin_a"), minimize=True),
Objective(metric=get_branin_metric(name="branin_b"), minimize=True),
]
if num_objectives == 3:
objectives.append(
Objective(metric=get_branin_metric(name="branin_c"), minimize=True)
)
return MultiObjective(objectives=objectives)
##############################
# Optimization Configs
##############################
[docs]
def get_optimization_config(
outcome_constraint: bool = True, relative: bool = True
) -> OptimizationConfig:
objective = get_objective()
outcome_constraints = (
[get_outcome_constraint(relative=relative)] if outcome_constraint else []
)
return OptimizationConfig(
objective=objective, outcome_constraints=outcome_constraints
)
[docs]
def get_map_optimization_config() -> OptimizationConfig:
objective = get_map_objective()
return OptimizationConfig(objective=objective)
[docs]
def get_multi_objective_optimization_config(
custom_metric: bool = False,
relative: bool = True,
outcome_constraint: bool = True,
) -> MultiObjectiveOptimizationConfig:
objective = get_custom_multi_objective() if custom_metric else get_multi_objective()
outcome_constraints = (
[get_outcome_constraint(relative=relative)] if outcome_constraint else []
)
objective_thresholds = [
get_objective_threshold(metric_name="m1"),
get_objective_threshold(metric_name="m3", comparison_op=ComparisonOp.LEQ),
]
return MultiObjectiveOptimizationConfig(
objective=objective,
outcome_constraints=outcome_constraints,
objective_thresholds=objective_thresholds,
)
[docs]
def get_optimization_config_no_constraints(
minimize: bool = False,
) -> OptimizationConfig:
return OptimizationConfig(
objective=Objective(metric=Metric("test_metric"), minimize=minimize)
)
[docs]
def get_branin_optimization_config(minimize: bool = False) -> OptimizationConfig:
return OptimizationConfig(objective=get_branin_objective(minimize=minimize))
def _validate_num_objectives(num_objectives: int) -> None:
if num_objectives not in (2, 3):
raise NotImplementedError("Only 2 and 3 objectives are supported.")
[docs]
def get_branin_multi_objective_optimization_config(
has_objective_thresholds: bool = False,
num_objectives: int = 2,
with_relative_constraint: bool = False,
with_absolute_constraint: bool = False,
) -> MultiObjectiveOptimizationConfig:
_validate_num_objectives(num_objectives=num_objectives)
# minimum Branin value is 0.397887
if has_objective_thresholds:
objective_thresholds = [
ObjectiveThreshold(
metric=get_branin_metric(name="branin_a"),
bound=10.0,
op=ComparisonOp.LEQ,
relative=False,
),
ObjectiveThreshold(
metric=get_branin_metric(name="branin_b"),
bound=20.0,
op=ComparisonOp.LEQ,
relative=False,
),
]
if num_objectives == 3:
objective_thresholds.append(
ObjectiveThreshold(
metric=get_branin_metric(name="branin_c"),
bound=5.0,
op=ComparisonOp.LEQ,
relative=False,
)
)
else:
objective_thresholds = None
outcome_constraints = []
if with_relative_constraint:
outcome_constraints.append(
get_outcome_constraint(
metric=get_branin_metric(name="branin_d"), relative=True
)
)
if with_absolute_constraint:
outcome_constraints.append(
get_outcome_constraint(
metric=get_branin_metric(name="branin_e"), relative=False
)
)
return MultiObjectiveOptimizationConfig(
objective=get_branin_multi_objective(num_objectives=num_objectives),
objective_thresholds=objective_thresholds,
outcome_constraints=outcome_constraints,
)
##############################
# Arms
##############################
[docs]
def get_arm() -> Arm:
# Expected `Dict[str, typing.Optional[typing.Union[bool, float, str]]]` for 2nd
# parameter `parameters` to call `ax.core.arm.Arm.__init__` but got
# `Dict[str, typing.Union[float, str]]`.
return Arm(parameters={"w": 0.75, "x": 1, "y": "foo", "z": True})
[docs]
def get_status_quo() -> Arm:
return Arm(
# Expected `Dict[str, typing.Optional[typing.Union[bool, float, str]]]` for 2nd
# parameter `parameters` to call `ax.core.arm.Arm.__init__`
# but got `Dict[str, typing.Union[float, str]]`.
parameters={"w": 0.2, "x": 1, "y": "bar", "z": False},
name="status_quo",
)
[docs]
def get_arm_weights1() -> MutableMapping[Arm, float]:
parameters_dicts: list[TParameterization] = [
{"w": 0.85, "x": 1, "y": "baz", "z": False},
{"w": 0.75, "x": 1, "y": "foo", "z": True},
{"w": 1.4, "x": 2, "y": "bar", "z": True},
]
arms = [Arm(param_dict) for param_dict in parameters_dicts]
weights = [0.25, 0.5, 0.25]
return OrderedDict(zip(arms, weights))
[docs]
def get_arm_weights2() -> MutableMapping[Arm, float]: # update
parameters_dicts: list[TParameterization] = [
{"w": 0.96, "x": 3, "y": "hello", "z": True},
{"w": 0.16, "x": 4, "y": "dear", "z": True},
{"w": 3.1, "x": 5, "y": "world", "z": False},
]
arms = [Arm(param_dict) for param_dict in parameters_dicts]
weights = [0.25, 0.5, 0.25]
return OrderedDict(zip(arms, weights))
[docs]
def get_arms_from_dict(arm_weights_dict: MutableMapping[Arm, float]) -> list[Arm]:
return list(arm_weights_dict.keys())
[docs]
def get_weights_from_dict(arm_weights_dict: MutableMapping[Arm, float]) -> list[float]:
return list(arm_weights_dict.values())
[docs]
def get_arms() -> list[Arm]:
return list(get_arm_weights1().keys())
[docs]
def get_weights() -> list[float]:
return list(get_arm_weights1().values())
[docs]
def get_branin_arms(n: int, seed: int) -> list[Arm]:
np.random.seed(seed)
x1_raw = np.random.rand(n)
x2_raw = np.random.rand(n)
return [
Arm(parameters={"x1": -5 + x1_raw[i] * 15, "x2": x2_raw[i] * 15})
for i in range(n)
]
[docs]
def get_abandoned_arm() -> AbandonedArm:
return AbandonedArm(name="0_0", reason="foobar", time=datetime.now())
##############################
# Generator Runs
##############################
[docs]
def get_generator_run() -> GeneratorRun:
arms = get_arms_from_dict(get_arm_weights1())
weights = get_weights_from_dict(get_arm_weights1())
optimization_config = get_optimization_config()
search_space = get_search_space()
arm_predictions = get_model_predictions_per_arm()
return GeneratorRun(
arms=arms,
weights=weights,
optimization_config=optimization_config,
search_space=search_space,
model_predictions=get_model_predictions(),
best_arm_predictions=(arms[0], arm_predictions[arms[0].signature]),
fit_time=10.0,
gen_time=5.0,
model_key="Sobol",
model_kwargs={"scramble": False, "torch_device": torch.device("cpu")},
bridge_kwargs={"transforms": Cont_X_trans, "torch_dtype": torch.double},
generation_step_index=0,
candidate_metadata_by_arm_signature={
a.signature: {"md_key": f"md_val_{a.signature}"} for a in arms
},
)
[docs]
def get_generator_run2() -> GeneratorRun:
arms = get_arms_from_dict(get_arm_weights1())
weights = get_weights_from_dict(get_arm_weights1())
return GeneratorRun(arms=arms, weights=weights)
##############################
# Runners
##############################
[docs]
def get_synthetic_runner() -> SyntheticRunner:
return SyntheticRunner(dummy_metadata="foobar")
##############################
# Data
##############################
[docs]
def get_data(
metric_name: str = "ax_test_metric",
trial_index: int = 0,
num_non_sq_arms: int = 4,
include_sq: bool = True,
) -> Data:
assert num_non_sq_arms < 5, "Only up to 4 arms currently handled."
arm_names = ["status_quo"] if include_sq else []
arm_names += [f"{trial_index}_{i}" for i in range(num_non_sq_arms)]
num_arms = num_non_sq_arms + 1 if include_sq else num_non_sq_arms
df_dict = {
"trial_index": trial_index,
"metric_name": metric_name,
"arm_name": arm_names,
"mean": [1, 3, 2, 2.25, 1.75][:num_arms],
"sem": [0, 0.5, 0.25, 0.40, 0.15][:num_arms],
"n": [100, 100, 100, 100, 100][:num_arms],
}
return Data(df=pd.DataFrame.from_records(df_dict))
[docs]
def get_non_monolithic_branin_moo_data() -> Data:
now = datetime.now()
return Data(
df=pd.DataFrame.from_records(
[
{
"arm_name": "status_quo",
"trial_index": "0",
"metric_name": "branin_a", # Obj. metric for experiment.
"mean": 2.0,
"sem": 0.01,
"start_time": now - timedelta(days=3),
"end_time": now,
},
{
"arm_name": "0_0",
"trial_index": "0",
"metric_name": "branin_a", # Obj. metric for experiment.
"mean": 1.0,
"sem": 0.01,
"start_time": now - timedelta(days=3),
"end_time": now,
},
{
"arm_name": "status_quo",
"trial_index": "0",
"metric_name": "branin_b", # Obj. metric for experiment.
"mean": 2.0,
"sem": 0.01,
"start_time": now - timedelta(days=2),
"end_time": now - timedelta(days=1),
},
{
"arm_name": "0_0",
"trial_index": "0",
"metric_name": "branin_b", # Obj. metric for experiment.
"mean": 1.0,
"sem": 0.01,
"start_time": now - timedelta(days=2),
"end_time": now - timedelta(days=1),
},
]
)
)
[docs]
def get_map_data(trial_index: int = 0) -> MapData:
evaluations = {
"status_quo": [
({"epoch": 1}, {"ax_test_metric": (1.0, 0.5)}),
({"epoch": 2}, {"ax_test_metric": (2.0, 0.5)}),
({"epoch": 3}, {"ax_test_metric": (3.0, 0.5)}),
({"epoch": 4}, {"ax_test_metric": (4.0, 0.5)}),
],
"0_0": [
({"epoch": 1}, {"ax_test_metric": (3.7, 0.5)}),
({"epoch": 2}, {"ax_test_metric": (3.8, 0.5)}),
({"epoch": 3}, {"ax_test_metric": (3.9, 0.5)}),
({"epoch": 4}, {"ax_test_metric": (4.0, 0.5)}),
],
"0_1": [
({"epoch": 1}, {"ax_test_metric": (3.0, 0.5)}),
({"epoch": 2}, {"ax_test_metric": (5.0, 0.5)}),
({"epoch": 3}, {"ax_test_metric": (6.0, 0.5)}),
({"epoch": 4}, {"ax_test_metric": (1.0, 0.5)}),
],
}
return MapData.from_map_evaluations(
evaluations=evaluations, # pyre-ignore [6]: Spurious param type mismatch.
trial_index=trial_index,
map_key_infos=[get_map_key_info()],
)
[docs]
def get_observations_with_invalid_value(invalid_value: float) -> list[Observation]:
obsd_with_non_finite = ObservationData(
metric_names=["m1"] * 4,
means=np.array([-100, 4, invalid_value, 2]),
covariance=np.eye(4),
)
observations = [
Observation(features=ObservationFeatures({}), data=obsd_with_non_finite)
]
return observations
# pyre-fixme[24]: Generic type `MapKeyInfo` expects 1 type parameter.
[docs]
def get_map_key_info() -> MapKeyInfo:
return MapKeyInfo(key="epoch", default_value=0.0)
[docs]
def get_branin_data(
trial_indices: Iterable[int] | None = None,
trials: Iterable[Trial] | None = None,
) -> Data:
if trial_indices and trials:
raise ValueError("Expected `trial_indices` or `trials`, not both.")
if trials:
df_dicts = [
{
"trial_index": trial.index,
"metric_name": "branin",
"arm_name": none_throws(checked_cast(Trial, trial).arm).name,
"mean": branin(
float(none_throws(none_throws(trial.arm).parameters["x1"])),
float(none_throws(none_throws(trial.arm).parameters["x2"])),
),
"sem": 0.0,
}
for trial in trials
]
else:
df_dicts = [
{
"trial_index": trial_index,
"metric_name": "branin",
"arm_name": f"{trial_index}_0",
"mean": 5.0,
"sem": 0.0,
}
for trial_index in (trial_indices or [0])
]
return Data(df=pd.DataFrame.from_records(df_dicts))
[docs]
def get_branin_data_batch(
batch: BatchTrial, fill_vals: dict[str, float] | None = None
) -> Data:
means = []
fill_vals = fill_vals or {}
for arm in batch.arms:
params = arm.parameters
for k, v in fill_vals.items():
if params.get(k, None) is None:
params[k] = v
if params["x1"] is None or params["x2"] is None:
means.append(5.0)
else:
means.append(
branin(
float(none_throws(params["x1"])),
float(none_throws(params["x2"])),
)
)
return Data(
pd.DataFrame(
{
"trial_index": batch.index,
"arm_name": [arm.name for arm in batch.arms],
"metric_name": "branin",
"mean": means,
"sem": 0.1,
}
)
)
[docs]
def get_branin_data_multi_objective(
trial_indices: Iterable[int] | None = None, num_objectives: int = 2
) -> Data:
_validate_num_objectives(num_objectives=num_objectives)
suffixes = ["a", "b"]
if num_objectives == 3:
suffixes.append("c")
df_dicts = [
{
"trial_index": trial_index,
"metric_name": f"branin_{suffix}",
"arm_name": f"{trial_index}_0",
"mean": 5.0,
"sem": 0.0,
}
for trial_index in (trial_indices or [0])
for suffix in suffixes
]
return Data(df=pd.DataFrame.from_records(df_dicts))
[docs]
def get_percentile_early_stopping_strategy() -> PercentileEarlyStoppingStrategy:
return PercentileEarlyStoppingStrategy(
percentile_threshold=0.25,
min_progression=0.2,
min_curves=10,
trial_indices_to_ignore=[0, 1, 2],
normalize_progressions=True,
)
[docs]
def get_percentile_early_stopping_strategy_with_non_objective_metric_name() -> (
PercentileEarlyStoppingStrategy
):
return PercentileEarlyStoppingStrategy(
metric_names=["foo"],
percentile_threshold=0.25,
min_progression=0.2,
min_curves=10,
trial_indices_to_ignore=[0, 1, 2],
normalize_progressions=True,
)
[docs]
def get_threshold_early_stopping_strategy() -> ThresholdEarlyStoppingStrategy:
return ThresholdEarlyStoppingStrategy(
metric_threshold=0.1,
min_progression=0.2,
trial_indices_to_ignore=[0, 1, 2],
normalize_progressions=True,
)
[docs]
def get_and_early_stopping_strategy() -> AndEarlyStoppingStrategy:
return AndEarlyStoppingStrategy(
left=get_percentile_early_stopping_strategy(),
right=get_threshold_early_stopping_strategy(),
)
[docs]
def get_or_early_stopping_strategy() -> OrEarlyStoppingStrategy:
return OrEarlyStoppingStrategy(
left=get_percentile_early_stopping_strategy(),
right=get_threshold_early_stopping_strategy(),
)
[docs]
class DummyEarlyStoppingStrategy(BaseEarlyStoppingStrategy):
def __init__(self, early_stop_trials: dict[int, str | None] | None = None) -> None:
self.early_stop_trials: dict[int, str | None] = early_stop_trials or {}
self.seconds_between_polls = 1
[docs]
def should_stop_trials_early(
self,
trial_indices: set[int],
experiment: Experiment,
**kwargs: dict[str, Any],
) -> dict[int, str | None]:
return self.early_stop_trials
[docs]
def get_improvement_global_stopping_strategy() -> ImprovementGlobalStoppingStrategy:
return ImprovementGlobalStoppingStrategy(
min_trials=30,
window_size=10,
improvement_bar=0.05,
inactive_when_pending_trials=True,
)
[docs]
class DummyGlobalStoppingStrategy(BaseGlobalStoppingStrategy):
"""
A dummy Global Stopping Strategy which stops the optimization after
a pre-specified number of trials are completed.
"""
def __init__(self, min_trials: int, trial_to_stop: int) -> None:
super().__init__(min_trials=min_trials)
self.trial_to_stop = trial_to_stop
def _should_stop_optimization(
self, experiment: Experiment, **kwargs: dict[str, Any]
) -> tuple[bool, str]:
num_completed_trials = len(experiment.trials_by_status[TrialStatus.COMPLETED])
if num_completed_trials >= max([self.min_trials, self.trial_to_stop]):
return True, "Stop the optimization."
else:
return False, ""
##############################
# Instances of types from core/types.py
##############################
[docs]
def get_model_mean() -> TModelMean:
mean: TModelMean = {"test_metric_1": [1, 2, 3], "test_metric_2": [3, 4, 5]}
return mean
[docs]
def get_model_covariance() -> TModelCov:
covariance: TModelCov = {
"test_metric_1": {"test_metric_1": [5, 6, 7], "test_metric_2": [7, 8, 9]},
"test_metric_2": {"test_metric_1": [9, 10, 11], "test_metric_2": [11, 12, 13]},
}
return covariance
[docs]
def get_model_predictions() -> TModelPredict:
model_predictions: TModelPredict = (get_model_mean(), get_model_covariance())
return model_predictions
[docs]
def get_model_predictions_per_arm() -> dict[str, TModelPredictArm]:
arms = list(get_arm_weights1().keys())
means = get_model_mean()
covariances = get_model_covariance()
metric_names = list(means.keys())
m_1, m_2 = metric_names[0], metric_names[1]
return {
arms[i].signature: (
{m_1: means[m_1][i], m_2: means[m_2][i]},
{
m_1: {m_1: covariances[m_1][m_1][i], m_2: covariances[m_1][m_2][i]},
m_2: {m_1: covariances[m_2][m_1][i], m_2: covariances[m_2][m_2][i]},
},
)
for i in range(len(arms))
}
##############################
# Modular BoTorch Model Components
##############################
[docs]
def get_botorch_model() -> BoTorchModel:
return BoTorchModel(
surrogate=get_surrogate(), acquisition_class=get_acquisition_type()
)
[docs]
def get_botorch_model_with_default_acquisition_class() -> BoTorchModel:
return BoTorchModel(
surrogate=get_surrogate(),
acquisition_class=Acquisition,
botorch_acqf_class=get_acquisition_function_type(),
)
[docs]
def get_botorch_model_with_surrogate_specs() -> BoTorchModel:
return BoTorchModel(
surrogate_specs={
"name": SurrogateSpec(botorch_model_kwargs={"some_option": "some_value"})
}
)
[docs]
def get_botorch_model_with_surrogate_spec() -> BoTorchModel:
return BoTorchModel(
surrogate_spec=SurrogateSpec(botorch_model_kwargs={"some_option": "some_value"})
)
[docs]
def get_surrogate() -> Surrogate:
return Surrogate(
botorch_model_class=get_model_type(),
mll_class=get_mll_type(),
)
[docs]
def get_surrogate_spec_with_default() -> SurrogateSpec:
return SurrogateSpec(
botorch_model_class=SingleTaskGP,
covar_module_class=ScaleMaternKernel,
covar_module_kwargs={
"ard_num_dims": DEFAULT,
"lengthscale_prior": GammaPrior(6.0, 3.0),
"outputscale_prior": GammaPrior(2.0, 0.15),
"batch_shape": DEFAULT,
},
)
[docs]
def get_surrogate_spec_with_lognormal() -> SurrogateSpec:
return SurrogateSpec(
botorch_model_class=SingleTaskGP,
covar_module_class=RBFKernel,
covar_module_kwargs={
"ard_num_dims": DEFAULT,
"lengthscale_prior": LogNormalPrior(-4.0, 1.0),
"batch_shape": DEFAULT,
},
)
[docs]
def get_acquisition_type() -> type[Acquisition]:
return Acquisition
[docs]
def get_model_type() -> type[Model]:
return SingleTaskGP
[docs]
def get_mll_type() -> type[MarginalLogLikelihood]:
return ExactMarginalLogLikelihood
[docs]
def get_acquisition_function_type() -> type[AcquisitionFunction]:
return qExpectedImprovement
[docs]
def get_sebo_acquisition_class() -> type[SEBOAcquisition]:
return SEBOAcquisition
[docs]
def get_winsorization_config() -> WinsorizationConfig:
return WinsorizationConfig(
lower_quantile_margin=0.2,
upper_quantile_margin=0.3,
lower_boundary=20,
upper_boundary=50,
)
[docs]
def get_gamma_prior() -> GammaPrior:
return GammaPrior(concentration=0.9, rate=10.0)
[docs]
def get_interval() -> Interval:
return Interval(lower_bound=1e-6, upper_bound=0.1)
##############################
# Scheduler
##############################
[docs]
def get_default_scheduler_options() -> SchedulerOptions:
return SchedulerOptions()
[docs]
def get_scheduler_options_batch_trial() -> SchedulerOptions:
return SchedulerOptions(trial_type=TrialType.BATCH_TRIAL)
##############################
# Other
##############################
[docs]
def get_risk_measure() -> RiskMeasure:
return RiskMeasure(risk_measure="Expectation", options={"n_w": 8})
[docs]
def get_parameter_distribution() -> ParameterDistribution:
return ParameterDistribution(
parameters=["x"],
distribution_class="norm",
distribution_parameters={"loc": 1.0, "scale": 0.5},
)
[docs]
def get_pathlib_path() -> Path:
return Path("some/meaningless/path")
[docs]
def get_dataset(
num_samples: int = 2,
d: int = 2,
m: int = 2,
has_observation_noise: bool = False,
feature_names: list[str] | None = None,
outcome_names: list[str] | None = None,
tkwargs: dict[str, Any] | None = None,
seed: int | None = None,
) -> SupervisedDataset:
"""Constructs a SupervisedDataset based on the given arguments.
Args:
num_samples: The number of samples in the dataset.
d: The dimension of the features.
m: The number of outcomes.
has_observation_noise: If True, includes Yvar in the dataset.
feature_names: A list of feature names. Defaults to x0, x1...
outcome_names: A list of outcome names. Defaults to y0, y1...
tkwargs: Optional dictionary of tensor kwargs, such as dtype and device.
seed: An optional seed used to generate the data.
"""
if seed is not None:
set_rng_seed(seed)
feature_names = feature_names or [f"x{i}" for i in range(d)]
outcome_names = outcome_names or [f"y{i}" for i in range(m)]
tkwargs = tkwargs or {}
return SupervisedDataset(
X=torch.rand(num_samples, d, **tkwargs),
Y=torch.rand(num_samples, m, **tkwargs),
Yvar=(
torch.rand(num_samples, m, **tkwargs) * 0.01
if has_observation_noise
else None
),
feature_names=feature_names,
outcome_names=outcome_names,
)
[docs]
def get_online_sobol_mbm_generation_strategy(
sobol_steps: int = 1,
) -> GenerationStrategy:
"""Constructs a GenerationStrategy with Sobol and MBM nodes for simulating
online optimization.
"""
# Set up the node-based generation strategy for testing.
# TODO: @mgarrard make this more realistic of an actual online gs
step_model_kwargs = {"silently_filter_kwargs": True}
sobol_criterion = [
MaxTrials(
threshold=1,
transition_to="MBM_node",
block_gen_if_met=True,
only_in_statuses=None,
not_in_statuses=[TrialStatus.FAILED, TrialStatus.ABANDONED],
),
MinTrials(
threshold=1,
transition_to="MBM_node",
block_gen_if_met=True,
only_in_statuses=[
TrialStatus.RUNNING,
TrialStatus.COMPLETED,
TrialStatus.EARLY_STOPPED,
],
),
]
sobol_model_spec = ModelSpec(
model_enum=Models.SOBOL,
model_kwargs=step_model_kwargs,
model_gen_kwargs={},
)
mbm_model_spec = ModelSpec(
model_enum=Models.BOTORCH_MODULAR,
model_kwargs=step_model_kwargs,
model_gen_kwargs={},
)
sobol_node = GenerationNode(
node_name="sobol_node",
transition_criteria=sobol_criterion,
model_specs=[sobol_model_spec],
input_constructors={InputConstructorPurpose.N: NodeInputConstructors.ALL_N},
)
mbm_node = GenerationNode(
node_name="MBM_node",
transition_criteria=[],
model_specs=[mbm_model_spec],
input_constructors={InputConstructorPurpose.N: NodeInputConstructors.ALL_N},
)
return GenerationStrategy(
name="Sobol+MBM_Nodes",
nodes=[sobol_node, mbm_node],
)
##############################
# Custom runner and metric
##############################
[docs]
class CustomTestRunner(Runner):
def __init__(self, test_attribute: str) -> None:
self.test_attribute = test_attribute
[docs]
def run(self, trial: BaseTrial) -> dict[str, Any]:
return {"foo": "bar"}
[docs]
class CustomTestMetric(Metric):
def __init__(
self,
name: str,
test_attribute: str,
lower_is_better: bool | None = None,
) -> None:
self.test_attribute = test_attribute
super().__init__(name=name, lower_is_better=lower_is_better)
[docs]
class SpecialGenerationStrategy(GenerationStrategyInterface):
"""A subclass of `GenerationStrategyInterface` to be used
for testing how methods respond to subtypes other than
`GenerationStrategy`."""
def __init__(self) -> None:
self._name = "special"
self._generator_runs: list[GeneratorRun] = []
[docs]
def gen_for_multiple_trials_with_multiple_models(
self,
experiment: Experiment,
data: Data | None = None,
pending_observations: dict[str, list[ObservationFeatures]] | None = None,
n: int | None = None,
num_trials: int = 1,
arms_per_node: dict[str, int] | None = None,
) -> list[list[GeneratorRun]]:
return []
[docs]
def clone_reset(self) -> SpecialGenerationStrategy:
clone = SpecialGenerationStrategy()
clone._name = self._name
return clone
