POMDPPlanners.tests.test_environments.test_light_dark_pomdp.light_dark_pomdp_utils package
Submodules
POMDPPlanners.tests.test_environments.test_light_dark_pomdp.light_dark_pomdp_utils.benchmark_reward_models module
Performance benchmark: light_dark env-level reward / reward_batch.
Times the public Environment-API reward methods
(ContinuousLightDarkPOMDP.reward() and
ContinuousLightDarkPOMDP.reward_batch()) end-to-end so the numbers
include the thin Python wrapper at continuous_light_dark_pomdp.py:637
on top of the reward model’s compute_reward / compute_reward_batch.
Covers all three reward-model variants (Standard, ZERO_MEAN_HAZARD_SHOCK,
DISTANCE_DECAYED_HAZARD_PENALTY) on a fixed workload. Used to compare BEFORE vs
AFTER the dangerous-area generic-kernel refactor.
Note: the C++ _native.simulate_rollout path is a separate hot path
with its own embedded reward logic and is not exercised here — to
benchmark that, time env._rollout or a full MCTS rollout instead.
- Run manually:
python -m POMDPPlanners.tests.test_environments.test_light_dark_pomdp.light_dark_pomdp_utils.benchmark_reward_models