def testGetProperties(self):
# Extract default value.
properties = serialize_init_args(Metric(name="foo"))
self.assertEqual(properties, {"name": "foo", "lower_is_better": None})
# Extract passed value.
properties = serialize_init_args(
Metric(name="foo", lower_is_better=True))
self.assertEqual(properties, {"name": "foo", "lower_is_better": True})
def serialize_init_args(cls, metric: "Metric") -> Dict[str, Any]:
"""Serialize the properties needed to initialize the metric.
Used for storage.
"""
return serialize_init_args(
object=metric,
exclude_fields=["name", "lower_is_better", "precomp_config"])
def serialize_init_args(cls, data: Data) -> Dict[str, Any]:
"""Serialize the class-dependent properties needed to initialize this Data.
Used for storage and to help construct new similar Data. All kwargs
other than "dataframe" and "description" are considered structural.
"""
return serialize_init_args(object=data,
exclude_fields=["df", "description"])
def benchmark_problem_to_dict(
benchmark_problem: BenchmarkProblem) -> Dict[str, Any]:
"""Converts an Ax benchmark problem to a serializable dictionary."""
if isinstance(benchmark_problem, SimpleBenchmarkProblem):
if benchmark_problem.uses_synthetic_function:
function_name = benchmark_problem.f.name # pyre-ignore[16]
f = None
else:
function_name = benchmark_problem.f.__name__ # pyre-ignore[16]
f = pickle.dumps(benchmark_problem.f, 0).decode()
return {
"__type": benchmark_problem.__class__.__name__,
"uses_synthetic_function":
benchmark_problem.uses_synthetic_function,
"function_name": function_name,
# If the benchamrk problem uses a custom callable, pickle it.
"f": f,
"name": benchmark_problem.name,
"domain": benchmark_problem.domain,
"minimize": benchmark_problem.minimize,
"noise_sd": benchmark_problem.noise_sd,
"evaluate_suggested": benchmark_problem.evaluate_suggested,
"optimal_value": benchmark_problem.optimal_value,
}
elif isinstance(benchmark_problem, BenchmarkProblem):
properties = serialize_init_args(object=benchmark_problem)
properties["__type"] = benchmark_problem.__class__.__name__
return properties
else: # pragma: no cover
raise ValueError(
f"Expected benchmark problem, got: {benchmark_problem}.")
def clone(self) -> "Runner":
"""Create a copy of this Runner."""
cls = type(self)
# pyre-ignore[45]: Cannot instantiate abstract class `Runner`.
return cls(
**serialize_init_args(self),
)
def clone(self) -> "Metric":
"""Create a copy of this Metric."""
cls = type(self)
return cls(**serialize_init_args(self), )
def runner_to_dict(runner: SyntheticRunner) -> Dict[str, Any]:
"""Convert Ax synthetic runner to a dictionary."""
properties = serialize_init_args(object=runner)
properties["__type"] = runner.__class__.__name__
return properties
def metric_to_dict(metric: Metric) -> Dict[str, Any]:
"""Convert Ax metric to a dictionary."""
properties = serialize_init_args(object=metric)
properties["__type"] = metric.__class__.__name__
return properties
def serialize_init_args(cls, runner: "Runner") -> Dict[str, Any]:
"""Serialize the properties needed to initialize the runner.
Used for storage.
"""
return serialize_init_args(object=runner)
def serialize_init_args(cls, metric: Metric) -> Dict[str, Any]:
"""Serialize the properties needed to initialize the metric.
Used for storage.
"""
return serialize_init_args(object=metric)
