def _transform_observation_data(
self, observation_data: List[ObservationData]
) -> Any: # TODO(jej): Make return type parametric
"""Apply terminal transform to given observation data and return result.
Converts a set of observation data to a tuple of
- an (n x m) array of means
- an (n x m x m) array of covariances
"""
try:
return observation_data_to_array(observation_data=observation_data)
except (KeyError, TypeError): # pragma: no cover
raise ValueError("Invalid formatting of observation data.")
def testObservationDataToArray(self):
outcomes = ["a", "b", "c"]
obsd = ObservationData(
metric_names=["c", "a", "b"],
means=np.array([1, 2, 3]),
covariance=np.array([[1, 2, 3], [4, 5, 6], [7, 8, 9]]),
)
Y, Ycov = observation_data_to_array(outcomes=outcomes,
observation_data=[obsd])
self.assertTrue(np.array_equal(Y, np.array([[2, 3, 1]])))
self.assertTrue(
np.array_equal(Ycov, np.array([[[5, 6, 4], [8, 9, 7], [2, 3,
1]]])))
def get_training_data(
self,
experiment: Experiment,
map_data: MapData,
keep_every_k_per_arm: Optional[int] = None,
) -> EarlyStoppingTrainingData:
"""Processes the raw (untransformed) training data into arrays for
use in modeling.
Args:
experiment: Experiment that contains the data.
map_data: The MapData from the experiment, as can be obtained by
via `_check_validity_and_get_data`.
keep_every_k_per_arm Subsample the learning curve by keeping every
kth entry. Useful for limiting training data for modeling.
Returns:
An `EarlyStoppingTrainingData` that contains training data arrays X, Y,
and Yvar + a list of arm names.
"""
if keep_every_k_per_arm is not None:
map_data = _subsample_map_data(
map_data=map_data, keep_every_k_per_arm=keep_every_k_per_arm)
observations = observations_from_map_data(experiment=experiment,
map_data=map_data,
map_keys_as_parameters=True)
obs_features, obs_data, arm_names = _unpack_observations(observations)
parameters = list(experiment.search_space.parameters.keys())
outcome = not_none(
experiment.optimization_config).objective.metric_names[0]
X = observation_features_to_array(parameters=parameters +
list(map_data.map_keys),
obsf=obs_features)
Y, Yvar = observation_data_to_array(outcomes=[outcome],
observation_data=obs_data)
return EarlyStoppingTrainingData(X=X,
Y=Y,
Yvar=Yvar,
arm_names=arm_names)