Skip to content

[Enhancement] Log mean/max KL across PPO update phase #523

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

Open
wants to merge 1 commit into
base: master
Choose a base branch
from
Open

[Enhancement] Log mean/max KL across PPO update phase #523

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
15 changes: 15 additions & 0 deletions cleanrl/ppo.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -241,6 +241,10 @@ def get_action_and_value(self, x, action=None):
# Optimizing the policy and value network
b_inds = np.arange(args.batch_size)
clipfracs = []

# tracking KL vals
old_kl_vals = []
kl_vals = []
for epoch in range(args.update_epochs):
np.random.shuffle(b_inds)
for start in range(0, args.batch_size, args.minibatch_size):
Expand All @@ -256,6 +260,8 @@ def get_action_and_value(self, x, action=None):
old_approx_kl = (-logratio).mean()
approx_kl = ((ratio - 1) - logratio).mean()
clipfracs += [((ratio - 1.0).abs() > args.clip_coef).float().mean().item()]
old_kl_vals.append(old_approx_kl)
kl_vals.append(approx_kl)

mb_advantages = b_advantages[mb_inds]
if args.norm_adv:
Expand Down Expand Up @@ -303,6 +309,15 @@ def get_action_and_value(self, x, action=None):
writer.add_scalar("losses/entropy", entropy_loss.item(), global_step)
writer.add_scalar("losses/old_approx_kl", old_approx_kl.item(), global_step)
writer.add_scalar("losses/approx_kl", approx_kl.item(), global_step)
# Aggregate KL stats across minibatches/epochs
if old_kl_vals:
_old_kl = torch.stack(old_kl_vals)
writer.add_scalar("losses/old_approx_kl_mean", _old_kl.mean().item(), global_step)
writer.add_scalar("losses/old_approx_kl_max", _old_kl.max().item(), global_step)
if kl_vals:
_kl = torch.stack(kl_vals)
writer.add_scalar("losses/approx_kl_mean", _kl.mean().item(), global_step)
writer.add_scalar("losses/approx_kl_max", _kl.max().item(), global_step)
writer.add_scalar("losses/clipfrac", np.mean(clipfracs), global_step)
writer.add_scalar("losses/explained_variance", explained_var, global_step)
print("SPS:", int(global_step / (time.time() - start_time)))
Expand Down
15 changes: 15 additions & 0 deletions cleanrl/ppo_continuous_action.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -256,6 +256,10 @@ def get_action_and_value(self, x, action=None):
# Optimizing the policy and value network
b_inds = np.arange(args.batch_size)
clipfracs = []

# tracking KL vals
old_kl_vals = []
kl_vals = []
for epoch in range(args.update_epochs):
np.random.shuffle(b_inds)
for start in range(0, args.batch_size, args.minibatch_size):
Expand All @@ -271,6 +275,8 @@ def get_action_and_value(self, x, action=None):
old_approx_kl = (-logratio).mean()
approx_kl = ((ratio - 1) - logratio).mean()
clipfracs += [((ratio - 1.0).abs() > args.clip_coef).float().mean().item()]
old_kl_vals.append(old_approx_kl)
kl_vals.append(approx_kl)

mb_advantages = b_advantages[mb_inds]
if args.norm_adv:
Expand Down Expand Up @@ -318,6 +324,15 @@ def get_action_and_value(self, x, action=None):
writer.add_scalar("losses/entropy", entropy_loss.item(), global_step)
writer.add_scalar("losses/old_approx_kl", old_approx_kl.item(), global_step)
writer.add_scalar("losses/approx_kl", approx_kl.item(), global_step)
# Aggregate KL stats across minibatches/epochs
if old_kl_vals:
_old_kl = torch.stack(old_kl_vals)
writer.add_scalar("losses/old_approx_kl_mean", _old_kl.mean().item(), global_step)
writer.add_scalar("losses/old_approx_kl_max", _old_kl.max().item(), global_step)
if kl_vals:
_kl = torch.stack(kl_vals)
writer.add_scalar("losses/approx_kl_mean", _kl.mean().item(), global_step)
writer.add_scalar("losses/approx_kl_max", _kl.max().item(), global_step)
writer.add_scalar("losses/clipfrac", np.mean(clipfracs), global_step)
writer.add_scalar("losses/explained_variance", explained_var, global_step)
print("SPS:", int(global_step / (time.time() - start_time)))
Expand Down