oxbrbtx/PrimAITE
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

#1386: remove setting of global seed + running pre-commit checks

Browse Source
This commit is contained in:
Czar Echavez
2023-07-06 12:10:26 +01:00
parent a35c363345
commit bb9bfc50a5
4 changed files with 9 additions and 27 deletions
Download Patch File Download Diff File Expand all files Collapse all files

22
tests/test_seeding_and_deterministic_session.py
View File

@@ -6,15 +6,11 @@ from tests import TEST_CONFIG_ROOT
@pytest.mark.parametrize(
"temp_primaite_session",
[[
TEST_CONFIG_ROOT / "ppo_seeded_training_config.yaml",
dos_very_basic_config_path()
]],
[[TEST_CONFIG_ROOT / "ppo_seeded_training_config.yaml", dos_very_basic_config_path()]],
indirect=True,
)
def test_seeded_learning(temp_primaite_session):
"""Test running seeded learning produces the same output when ran twice."""
expected_mean_reward_per_episode = {
1: -90.703125,
2: -91.15234375,
@@ -25,26 +21,22 @@ def test_seeded_learning(temp_primaite_session):
7: -88.984375,
8: -88.3203125,
9: -112.79296875,
10: -100.01953125
10: -100.01953125,
}
with temp_primaite_session as session:
assert session._training_config.seed == 67890, \
"Expected output is based upon a agent that was trained with " \
"seed 67890"
assert session._training_config.seed == 67890, (
"Expected output is based upon a agent that was trained with " "seed 67890"
)
session.learn()
actual_mean_reward_per_episode = session.learn_av_reward_per_episode()
assert actual_mean_reward_per_episode == expected_mean_reward_per_episode
@pytest.mark.skip(reason="Inconsistent results. Needs someone with RL "
"knowledge to investigate further.")
@pytest.mark.skip(reason="Inconsistent results. Needs someone with RL " "knowledge to investigate further.")
@pytest.mark.parametrize(
"temp_primaite_session",
[[
TEST_CONFIG_ROOT / "ppo_seeded_training_config.yaml",
dos_very_basic_config_path()
]],
[[TEST_CONFIG_ROOT / "ppo_seeded_training_config.yaml", dos_very_basic_config_path()]],
indirect=True,
)
def test_deterministic_evaluation(temp_primaite_session):