Skip to content

Retain final benchmark observations at simulated completion #5315

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

Closed
ltiao wants to merge 1 commit into from
Closed

Retain final benchmark observations at simulated completion #5315

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
10 changes: 4 additions & 6 deletions ax/benchmark/benchmark_metric.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -3,8 +3,6 @@
# This source code is licensed under the MIT license found in the
# LICENSE file in the root directory of this source tree.

# pyre-strict

"""
Metric classes for Ax benchmarking.

Expand Down Expand Up @@ -183,7 +181,7 @@ def fetch_trial_data(self, trial: BaseTrial, **kwargs: Any) -> MetricFetchResult
start_time = none_throws(sim_trial.sim_start_time)

if sim_trial.sim_completed_time is None: # Still running
max_t = backend_simulator.time - start_time
max_t = backend_simulator.time
elif sim_trial.sim_completed_time > backend_simulator.time:
raise RuntimeError(
"The trial's completion time is in the future! This is "
Expand All @@ -192,10 +190,10 @@ def fetch_trial_data(self, trial: BaseTrial, **kwargs: Any) -> MetricFetchResult
)
else:
# Completed, may have stopped early -- can't assume all data available
completed_time = none_throws(sim_trial.sim_completed_time)
max_t = completed_time - start_time
max_t = none_throws(sim_trial.sim_completed_time)

available_data = df[df["virtual runtime"] <= max_t]
# Compare absolute times to avoid cancellation when subtracting the start.
available_data = df[start_time + df["virtual runtime"] <= max_t]

if not self.observe_noise_sd:
available_data.loc[:, "sem"] = np.nan
Expand Down
55 changes: 53 additions & 2 deletions ax/benchmark/tests/test_benchmark_metric.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -3,8 +3,6 @@
# 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 math
from itertools import product
from unittest.mock import Mock
Expand Down Expand Up @@ -347,6 +345,59 @@ def test_fetch_trial_multiple_time_steps_with_simulator(self) -> None:
self._test_fetch_trial_multiple_time_steps_with_simulator(batch=False)
self._test_fetch_trial_multiple_time_steps_with_simulator(batch=True)

def test_fetch_trial_data_at_simulated_time_boundary(self) -> None:
start_time = 192.0
runtime = 320.10965660870767
completed_time = start_time + runtime
self.assertLess(completed_time - start_time, runtime)
for early_stop, metrics in product(
[False, True], [self.map_metrics, self.tv_metrics]
):
metric = metrics["test_metric1"]
with self.subTest(early_stop=early_stop, metric_cls=type(metric)):
trial = get_test_trial(multiple_time_steps=True)
metadata = trial.run_metadata["benchmark_metadata"]
df = metadata.dfs["test_metric1"]
df["virtual runtime"] = [0.0, runtime / 2, runtime]
df["mean"] = [0.0, 1.0, 2.0]
simulator = BackendSimulator(
options=BackendSimulatorOptions(internal_clock=start_time),
verbose_logging=False,
)
simulator.run_trial(trial.index, runtime=runtime)
trial.update_run_metadata(
{
"benchmark_metadata": BenchmarkTrialMetadata(
dfs=metadata.dfs, backend_simulator=simulator
)
}
)
simulator.options.internal_clock = np.nextafter(completed_time, -np.inf)
data = assert_is_instance(metric.fetch_trial_data(trial).value, Data)
self.assertEqual(data.df["mean"].tolist(), [1.0])
if isinstance(metric, MapMetric):
self.assertEqual(data.full_df["step"].tolist(), [0, 1])
if early_stop:
simulator.stop_trial(trial.index)
else:
simulator.options.internal_clock = completed_time
simulator._update_completed(completed_time)
data = assert_is_instance(metric.fetch_trial_data(trial).value, Data)
self.assertEqual(data.df["mean"].tolist(), [1.0 if early_stop else 2.0])
if isinstance(metric, MapMetric):
self.assertEqual(
data.full_df["step"].tolist(),
[0, 1] if early_stop else [0, 1, 2],
)
simulator.options.internal_clock = completed_time + 1
data = assert_is_instance(metric.fetch_trial_data(trial).value, Data)
self.assertEqual(data.df["mean"].tolist(), [1.0 if early_stop else 2.0])
if isinstance(metric, MapMetric):
self.assertEqual(
data.full_df["step"].tolist(),
[0, 1] if early_stop else [0, 1, 2],
)

def test_sim_trial_completes_in_future_raises(self) -> None:
simulator = BackendSimulator()
simulator.run_trial(trial_index=0, runtime=0)
Expand Down
Loading