huangfu/InteractionImitation
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Use collision rate as main metric

Browse Source
This commit is contained in:
ebuehrle
2022-02-26 14:28:58 +01:00
parent 1e24612347
commit 35e6fb299c
2 changed files with 10 additions and 10 deletions
Download Patch File Download Diff File Expand all files Collapse all files

12
sgail-ppo-options-setobs2.py
View File

@@ -92,20 +92,16 @@ def training_function(config):
expert_data = (torch.cat(d0), torch.cat(d1), torch.cat(d2), torch.cat(d3)) expert_data = (torch.cat(d0), torch.cat(d1), torch.cat(d2), torch.cat(d3))
expert_data = Buffer(*expert_data) expert_data = Buffer(*expert_data)
run_folder = str(datetime.now())
os.mkdir(os.path.join(DIR, run_folder))
with open(os.path.join(DIR, run_folder, 'config.json'), 'w') as f:
json.dump(config, f, indent=4)
def callback(info): def callback(info):
tune.report(gen_mean_reward_per_episode=info['gen/mean_reward_per_episode'], tune.report(gen_mean_reward_per_episode=info['gen/mean_reward_per_episode'],
disc_mean_reward_per_episode=info['disc/mean_reward_per_episode'], disc_mean_reward_per_episode=info['disc/mean_reward_per_episode'],
mean_episode_length=info['gen/mean_episode_length']) mean_episode_length=info['gen/mean_episode_length'],
gen_collision_rate=info['gen/collision_rate'])
# save model checkpoints # save model checkpoints
ep = info['epoch'] + 1 ep = info['epoch'] + 1
if (ep % 25 == 0): if (ep % 25 == 0):
torch.save(info['policy'].state_dict(), os.path.join(DIR, run_folder, f'policy_epoch{ep}.pt')) torch.save(info['policy'].state_dict(), f'policy_epoch{ep}.pt')
value, policy = gail_ppo( value, policy = gail_ppo(
env_fn=env_fn, env_fn=env_fn,
@@ -164,7 +160,7 @@ analysis = tune.run(
} }
) )
print('Best config: ', analysis.get_best_config(metric='gen_mean_reward_per_episode', mode='max')) print('Best config: ', analysis.get_best_config(metric='gen_collision_rate', mode='min'))
# %% # %%
# policy = SetMaskedDiscretePolicy(env_fn(0).action_space.n) # policy = SetMaskedDiscretePolicy(env_fn(0).action_space.n)

8
src/safe_options/options.py
View File

@@ -57,7 +57,8 @@ def gail(env_fn, expert_data, discriminator, disc_opt, disc_iters, policy, value
gen_mean_reward_per_episode = generator_data.hl.rewards[~generator_data.hl.dones].sum() / generator_data.hl.states.shape[0] gen_mean_reward_per_episode = generator_data.hl.rewards[~generator_data.hl.dones].sum() / generator_data.hl.states.shape[0]
logger.add_scalar('gen/mean_reward_per_episode', gen_mean_reward_per_episode, epoch) logger.add_scalar('gen/mean_reward_per_episode', gen_mean_reward_per_episode, epoch)
logger.add_scalar('gen/unsafe_probability_mass', policy.unsafe_probability_mass(policy(generator_data.hl.states[~generator_data.hl.dones], generator_data.hl.safe_actions[~generator_data.hl.dones])).mean(), epoch) logger.add_scalar('gen/unsafe_probability_mass', policy.unsafe_probability_mass(policy(generator_data.hl.states[~generator_data.hl.dones], generator_data.hl.safe_actions[~generator_data.hl.dones])).mean(), epoch)
logger.add_scalar('gen/collision_rate', (1. * collisions.any(-1)).mean(), epoch) gen_collision_rate = (1. * collisions.any(-1)).mean()
logger.add_scalar('gen/collision_rate', gen_collision_rate, epoch)
discriminator, loss = train_discriminator(expert_data, generator_data.ll, discriminator, disc_opt, disc_iters, wasserstein, wasserstein_c) discriminator, loss = train_discriminator(expert_data, generator_data.ll, discriminator, disc_opt, disc_iters, wasserstein, wasserstein_c)
if wasserstein: if wasserstein:
@@ -81,6 +82,7 @@ def gail(env_fn, expert_data, discriminator, disc_opt, disc_iters, policy, value
'policy': policy, 'policy': policy,
'gen/mean_episode_length': gen_mean_episode_length.item(), 'gen/mean_episode_length': gen_mean_episode_length.item(),
'gen/mean_reward_per_episode': gen_mean_reward_per_episode.item(), 'gen/mean_reward_per_episode': gen_mean_reward_per_episode.item(),
'gen/collision_rate': gen_collision_rate.item(),
'disc/mean_reward_per_episode': disc_mean_reward_per_episode.item(), 'disc/mean_reward_per_episode': disc_mean_reward_per_episode.item(),
}) })
@@ -103,7 +105,8 @@ def gail_ppo(env_fn, expert_data, discriminator, disc_opt, disc_iters, policy, v
gen_mean_reward_per_episode = generator_data.hl.rewards[~generator_data.hl.dones].sum() / generator_data.hl.states.shape[0] gen_mean_reward_per_episode = generator_data.hl.rewards[~generator_data.hl.dones].sum() / generator_data.hl.states.shape[0]
logger.add_scalar('gen/mean_reward_per_episode', gen_mean_reward_per_episode, epoch) logger.add_scalar('gen/mean_reward_per_episode', gen_mean_reward_per_episode, epoch)
logger.add_scalar('gen/unsafe_probability_mass', policy.unsafe_probability_mass(policy(generator_data.hl.states[~generator_data.hl.dones], generator_data.hl.safe_actions[~generator_data.hl.dones])).mean(), epoch) logger.add_scalar('gen/unsafe_probability_mass', policy.unsafe_probability_mass(policy(generator_data.hl.states[~generator_data.hl.dones], generator_data.hl.safe_actions[~generator_data.hl.dones])).mean(), epoch)
logger.add_scalar('gen/collision_rate', (1. * collisions.any(-1)).mean(), epoch) gen_collision_rate = (1. * collisions.any(-1)).mean()
logger.add_scalar('gen/collision_rate', gen_collision_rate, epoch)
discriminator, loss = train_discriminator(expert_data, generator_data.ll, discriminator, disc_opt, disc_iters, wasserstein, wasserstein_c) discriminator, loss = train_discriminator(expert_data, generator_data.ll, discriminator, disc_opt, disc_iters, wasserstein, wasserstein_c)
if wasserstein: if wasserstein:
@@ -127,6 +130,7 @@ def gail_ppo(env_fn, expert_data, discriminator, disc_opt, disc_iters, policy, v
'policy': policy, 'policy': policy,
'gen/mean_episode_length': gen_mean_episode_length.item(), 'gen/mean_episode_length': gen_mean_episode_length.item(),
'gen/mean_reward_per_episode': gen_mean_reward_per_episode.item(), 'gen/mean_reward_per_episode': gen_mean_reward_per_episode.item(),
'gen/collision_rate': gen_collision_rate.item(),
'disc/mean_reward_per_episode': disc_mean_reward_per_episode.item(), 'disc/mean_reward_per_episode': disc_mean_reward_per_episode.item(),
}) })