Skip to content
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

Jump to bottom

SQIL and PC performance check fixes #811

Merged
merged 5 commits into from
Oct 22, 2023
Merged

SQIL and PC performance check fixes #811

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
10f2a16
Reduce training times in SQIL tests to make the test suite faster.
ernestum Oct 9, 2023
8ffca91
Skip the continous SQIL test with TD3 since it is unstable.
ernestum Oct 9, 2023
cf29284
Disable SQIL test because it is flaky and slow.
ernestum Oct 9, 2023
f053932
Make SQIL tests more deterministic by adding more seeding.
ernestum Oct 9, 2023
1b20336
Increase training time and numer of samples to compare in the PC perf…
ernestum Oct 11, 2023
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
6 changes: 3 additions & 3 deletions tests/algorithms/test_preference_comparisons.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1068,7 +1068,7 @@ def test_that_trainer_improves(
novice_agent_rewards, _ = evaluation.evaluate_policy(
agent_trainer.algorithm.policy,
action_is_reward_venv,
25,
50,
return_episode_rewards=True,
)

Expand All @@ -1077,7 +1077,7 @@ def test_that_trainer_improves(
# after this training, and thus `later_rewards` should have lower loss.
first_reward_network_stats = main_trainer.train(20, 20)

later_reward_network_stats = main_trainer.train(50, 20)
later_reward_network_stats = main_trainer.train(100, 40)
assert (
first_reward_network_stats["reward_loss"]
> later_reward_network_stats["reward_loss"]
Expand All @@ -1087,7 +1087,7 @@ def test_that_trainer_improves(
trained_agent_rewards, _ = evaluation.evaluate_policy(
agent_trainer.algorithm.policy,
action_is_reward_venv,
25,
50,
return_episode_rewards=True,
)

Expand Down
12 changes: 7 additions & 5 deletions tests/algorithms/test_sqil.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -90,7 +90,7 @@ def _test_sqil_no_crash(
rl_algo_class=rl_algo_class,
rl_kwargs=rl_kwargs,
)
model.train(total_timesteps=5000)
model.train(total_timesteps=500)


def test_sqil_no_crash_discrete(
Expand All @@ -104,7 +104,7 @@ def test_sqil_no_crash_discrete(
cartpole_venv,
"seals/CartPole-v0",
rl_algo_class=dqn.DQN,
rl_kwargs=dict(learning_starts=1000),
rl_kwargs=dict(learning_starts=100),
)


Expand Down Expand Up @@ -143,7 +143,7 @@ def _test_sqil_few_demonstrations(
rl_algo_class=rl_algo_class,
rl_kwargs=rl_kwargs,
)
model.train(total_timesteps=1_000)
model.train(total_timesteps=1_00)


def test_sqil_few_demonstrations_discrete(
Expand All @@ -157,7 +157,7 @@ def test_sqil_few_demonstrations_discrete(
cartpole_venv,
"seals/CartPole-v0",
rl_algo_class=dqn.DQN,
rl_kwargs=dict(learning_starts=10),
rl_kwargs=dict(learning_starts=10, seed=42),
)


Expand All @@ -174,6 +174,7 @@ def test_sqil_few_demonstrations_continuous(
pendulum_single_venv,
"Pendulum-v1",
rl_algo_class=rl_algo_class,
rl_kwargs=dict(seed=42),
)


Expand Down Expand Up @@ -203,7 +204,7 @@ def _test_sqil_performance(
return_episode_rewards=True,
)

model.train(total_timesteps=10_000)
model.train(total_timesteps=1_000)

venv.seed(SEED)
rewards_after, _ = evaluate_policy(
Expand Down Expand Up @@ -239,6 +240,7 @@ def test_sqil_performance_discrete(
)


@pytest.mark.skip(reason="This test is flaky.")
@pytest.mark.parametrize("rl_algo_class", RL_ALGOS_CONT_ACTIONS)
def test_sqil_performance_continuous(
rng: np.random.Generator,
Expand Down
Loading