Skip to content

Preserve missing metrics when merging repeated measurements #5313

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
DiadkoShmek wants to merge 1 commit into
base: main
Choose a base branch
from
Open

Preserve missing metrics when merging repeated measurements #5313

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 2 additions & 0 deletions ax/adapter/transforms/merge_repeated_measurements.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -61,6 +61,8 @@ def __init__(
df_m = df[[("mean", m), ("sem", m)]].dropna(
axis=0, subset=[("mean", m)]
)
if df_m.empty:
continue
if any(df_m[("sem", m)].isna()):
raise NotImplementedError(
"All metrics must have noise observations."
Expand Down
65 changes: 65 additions & 0 deletions ax/adapter/transforms/tests/test_merge_repeated_measurements_transform.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -176,3 +176,68 @@ def test_with_experiment_data(self) -> None:
],
)
assert_frame_equal(transformed_data.observation_data, expected_obs_data)

def test_missing_metrics_remain_missing(self) -> None:
for missing_metric in ("m1", "m2"):
with self.subTest(missing_metric=missing_metric):
missing_index = ("m1", "m2").index(missing_metric)
means = [[2.0, 2.0], [4.0, 4.0], [6.0, 9.0]]
sems = [[1.0, 1.0], [2.0, 2.0], [3.0, 3.0]]
for i in (0, 1):
means[i][missing_index] = np.nan
sems[i][missing_index] = np.nan
experiment = get_experiment_with_observations(
observations=means,
sems=sems,
parameterizations=[{"x": 0.0}, {"x": 0.0}, {"x": 1.0}],
)
experiment_data = extract_experiment_data(
experiment=experiment, data_loader_config=DataLoaderConfig()
)
transformed_data = MergeRepeatedMeasurements(
experiment_data=experiment_data
).transform_experiment_data(deepcopy(experiment_data))
observations = transformed_data.convert_to_list_of_observations()
observed_metric = "m2" if missing_metric == "m1" else "m1"
self.assertTrue(
np.isnan(
transformed_data.observation_data.loc[
(0, "0_0"), ("mean", missing_metric)
]
)
)
self.assertTrue(
np.isnan(
transformed_data.observation_data.loc[
(0, "0_0"), ("sem", missing_metric)
]
)
)
self.assertEqual(
observations[0].data.metric_signatures, [observed_metric]
)
np.testing.assert_allclose(observations[0].data.means, [2.4])
np.testing.assert_allclose(observations[0].data.covariance, [[0.8]])
self.assertEqual(observations[1].data.metric_signatures, ["m1", "m2"])
np.testing.assert_allclose(observations[1].data.means, [6.0, 9.0])
np.testing.assert_allclose(
observations[1].data.covariance, np.diag([9.0, 9.0])
)

def test_metric_observed_only_in_later_trial(self) -> None:
experiment = get_experiment_with_observations(
observations=[[2.0, np.nan], [4.0, 8.0]],
sems=[[1.0, np.nan], [2.0, 2.0]],
parameterizations=[{"x": 0.0}, {"x": 0.0}],
)
experiment_data = extract_experiment_data(
experiment=experiment, data_loader_config=DataLoaderConfig()
)
transformed_data = MergeRepeatedMeasurements(
experiment_data=experiment_data
).transform_experiment_data(deepcopy(experiment_data))
observations = transformed_data.convert_to_list_of_observations()
self.assertEqual(len(observations), 1)
self.assertEqual(observations[0].data.metric_signatures, ["m1", "m2"])
np.testing.assert_allclose(observations[0].data.means, [2.4, 8.0])
np.testing.assert_allclose(observations[0].data.covariance, np.diag([0.8, 4.0]))