# 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 copy
from collections.abc import Callable, Mapping
from dataclasses import dataclass
from typing import Any
from ax.benchmark.benchmark_problem import BenchmarkProblem, create_problem_from_botorch
from ax.benchmark.problems.hd_embedding import embed_higher_dimension
from ax.benchmark.problems.hpo.torchvision import (
get_pytorch_cnn_torchvision_benchmark_problem,
)
from ax.benchmark.problems.runtime_funcs import int_from_params
from ax.benchmark.problems.synthetic.discretized.mixed_integer import (
get_discrete_ackley,
get_discrete_hartmann,
get_discrete_rosenbrock,
)
from ax.benchmark.problems.synthetic.hss.jenatton import get_jenatton_benchmark_problem
from botorch.test_functions import synthetic
from botorch.test_functions.multi_objective import BraninCurrin
[docs]
@dataclass
class BenchmarkProblemRegistryEntry:
factory_fn: Callable[..., BenchmarkProblem]
factory_kwargs: dict[str, Any]
# Baseline values were obtained with `compute_baseline_value_from_sobol`
BENCHMARK_PROBLEM_REGISTRY = {
"ackley4": BenchmarkProblemRegistryEntry(
factory_fn=create_problem_from_botorch,
factory_kwargs={
"test_problem_class": synthetic.Ackley,
"test_problem_kwargs": {"dim": 4},
"num_trials": 40,
"observe_noise_sd": False,
},
),
"ackley4_async_noisy": BenchmarkProblemRegistryEntry(
factory_fn=create_problem_from_botorch,
factory_kwargs={
"test_problem_class": synthetic.Ackley,
"test_problem_kwargs": {"dim": 4},
"num_trials": 40,
"noise_std": 1.0,
"observe_noise_sd": False,
"step_runtime_function": int_from_params,
"name": "ackley4_async_noisy",
},
),
"branin": BenchmarkProblemRegistryEntry(
factory_fn=create_problem_from_botorch,
factory_kwargs={
"test_problem_class": synthetic.Branin,
"test_problem_kwargs": {},
"num_trials": 30,
"observe_noise_sd": False,
},
),
"branin_currin": BenchmarkProblemRegistryEntry(
factory_fn=create_problem_from_botorch,
factory_kwargs={
"test_problem_class": BraninCurrin,
"test_problem_kwargs": {},
"num_trials": 30,
"observe_noise_sd": False,
},
),
"branin_currin30": BenchmarkProblemRegistryEntry(
factory_fn=lambda n, num_trials: embed_higher_dimension(
problem=create_problem_from_botorch(
test_problem_class=BraninCurrin,
test_problem_kwargs={},
num_trials=num_trials,
observe_noise_sd=False,
baseline_value=3.0187520516793587,
),
total_dimensionality=n,
),
factory_kwargs={"n": 30, "num_trials": 30},
),
"griewank4": BenchmarkProblemRegistryEntry(
factory_fn=create_problem_from_botorch,
factory_kwargs={
"test_problem_class": synthetic.Griewank,
"test_problem_kwargs": {"dim": 4},
"num_trials": 40,
"observe_noise_sd": False,
},
),
"hartmann3": BenchmarkProblemRegistryEntry(
factory_fn=create_problem_from_botorch,
factory_kwargs={
"test_problem_class": synthetic.Hartmann,
"test_problem_kwargs": {"dim": 3},
"num_trials": 30,
"observe_noise_sd": False,
},
),
"hartmann6": BenchmarkProblemRegistryEntry(
factory_fn=create_problem_from_botorch,
factory_kwargs={
"test_problem_class": synthetic.Hartmann,
"test_problem_kwargs": {"dim": 6},
"num_trials": 35,
"observe_noise_sd": False,
},
),
"hartmann30": BenchmarkProblemRegistryEntry(
factory_fn=lambda n, num_trials: embed_higher_dimension(
problem=create_problem_from_botorch(
test_problem_class=synthetic.Hartmann,
test_problem_kwargs={"dim": 6},
num_trials=num_trials,
observe_noise_sd=False,
),
total_dimensionality=n,
),
factory_kwargs={"n": 30, "num_trials": 25},
),
"hpo_pytorch_cnn_MNIST": BenchmarkProblemRegistryEntry(
factory_fn=get_pytorch_cnn_torchvision_benchmark_problem,
factory_kwargs={
"name": "MNIST",
"num_trials": 20,
},
),
"hpo_pytorch_cnn_FashionMNIST": BenchmarkProblemRegistryEntry(
factory_fn=get_pytorch_cnn_torchvision_benchmark_problem,
factory_kwargs={
"name": "FashionMNIST",
"num_trials": 50,
},
),
"jenatton": BenchmarkProblemRegistryEntry(
factory_fn=get_jenatton_benchmark_problem,
factory_kwargs={"num_trials": 50, "observe_noise_sd": False},
),
"levy4": BenchmarkProblemRegistryEntry(
factory_fn=create_problem_from_botorch,
factory_kwargs={
"test_problem_class": synthetic.Levy,
"test_problem_kwargs": {"dim": 4},
"num_trials": 40,
"observe_noise_sd": False,
},
),
"powell4": BenchmarkProblemRegistryEntry(
factory_fn=create_problem_from_botorch,
factory_kwargs={
"test_problem_class": synthetic.Powell,
"test_problem_kwargs": {"dim": 4},
"num_trials": 40,
"observe_noise_sd": False,
},
),
"rosenbrock4": BenchmarkProblemRegistryEntry(
factory_fn=create_problem_from_botorch,
factory_kwargs={
"test_problem_class": synthetic.Rosenbrock,
"test_problem_kwargs": {"dim": 4},
"num_trials": 40,
"observe_noise_sd": False,
},
),
"six_hump_camel": BenchmarkProblemRegistryEntry(
factory_fn=create_problem_from_botorch,
factory_kwargs={
"test_problem_class": synthetic.SixHumpCamel,
"test_problem_kwargs": {},
"num_trials": 30,
"observe_noise_sd": False,
},
),
"three_hump_camel": BenchmarkProblemRegistryEntry(
factory_fn=create_problem_from_botorch,
factory_kwargs={
"test_problem_class": synthetic.ThreeHumpCamel,
"test_problem_kwargs": {},
"num_trials": 30,
"observe_noise_sd": False,
},
),
# Problems where we observe the noise level
"branin_observed_noise": BenchmarkProblemRegistryEntry(
factory_fn=create_problem_from_botorch,
factory_kwargs={
"test_problem_class": synthetic.Branin,
"test_problem_kwargs": {},
"num_trials": 20,
"observe_noise_sd": True,
},
),
"branin_currin_observed_noise": BenchmarkProblemRegistryEntry(
factory_fn=create_problem_from_botorch,
factory_kwargs={
"test_problem_class": BraninCurrin,
"test_problem_kwargs": {},
"num_trials": 30,
"observe_noise_sd": True,
},
),
"branin_currin30_observed_noise": BenchmarkProblemRegistryEntry(
factory_fn=lambda n, num_trials: embed_higher_dimension(
problem=create_problem_from_botorch(
test_problem_class=BraninCurrin,
test_problem_kwargs={},
num_trials=num_trials,
observe_noise_sd=True,
),
total_dimensionality=n,
),
factory_kwargs={"n": 30, "num_trials": 30},
),
"hartmann6_observed_noise": BenchmarkProblemRegistryEntry(
factory_fn=create_problem_from_botorch,
factory_kwargs={
"test_problem_class": synthetic.Hartmann,
"test_problem_kwargs": {"dim": 6},
"num_trials": 50,
"observe_noise_sd": True,
},
),
"hartmann30_observed_noise": BenchmarkProblemRegistryEntry(
factory_fn=lambda n, num_trials: embed_higher_dimension(
problem=create_problem_from_botorch(
test_problem_class=synthetic.Hartmann,
test_problem_kwargs={"dim": 6},
num_trials=num_trials,
observe_noise_sd=True,
),
total_dimensionality=n,
),
factory_kwargs={"n": 30, "num_trials": 25},
),
"jenatton_observed_noise": BenchmarkProblemRegistryEntry(
factory_fn=get_jenatton_benchmark_problem,
factory_kwargs={"num_trials": 25, "observe_noise_sd": True},
),
"constrained_gramacy_observed_noise": BenchmarkProblemRegistryEntry(
factory_fn=create_problem_from_botorch,
factory_kwargs={
"test_problem_class": synthetic.ConstrainedGramacy,
"test_problem_kwargs": {},
"num_trials": 50,
"observe_noise_sd": True,
},
),
"Discrete Hartmann": BenchmarkProblemRegistryEntry(
factory_fn=get_discrete_hartmann,
factory_kwargs={},
),
"Discrete Ackley": BenchmarkProblemRegistryEntry(
factory_fn=get_discrete_ackley, factory_kwargs={}
),
"Discrete Rosenbrock": BenchmarkProblemRegistryEntry(
factory_fn=get_discrete_rosenbrock, factory_kwargs={}
),
}
[docs]
def get_problem(
problem_key: str,
registry: Mapping[str, BenchmarkProblemRegistryEntry] | None = None,
**additional_kwargs: Any,
) -> BenchmarkProblem:
"""
Generate a benchmark problem from a key, registry, and additional arguments.
Args:
problem_key: The key by which a `BenchmarkProblemRegistryEntry` is
looked up in the registry; a problem will then be generated from
that entry and `additional_kwargs`. Note that this is not
necessarily the same as the `name` attribute of the problem, and
that one `problem_key` can generate several different
`BenchmarkProblem`s by passing `additional_kwargs`. However, it is a
good practice to maintain a 1:1 mapping between `problem_key` and
the name.
registry: If not provided, uses `BENCHMARK_PROBLEM_REGISTRY` to use
problems defined within Ax.
additional_kwargs: Additional kwargs to pass to the factory function of
the `BenchmarkProblemRegistryEntry`.
"""
registry = BENCHMARK_PROBLEM_REGISTRY if registry is None else registry
entry = registry[problem_key]
kwargs = copy.copy(entry.factory_kwargs)
kwargs.update(additional_kwargs)
return entry.factory_fn(**kwargs)
