Skip to content

feature: compute rho, c by joint probability #322

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

feature: compute rho, c by joint probability #322

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
9 changes: 4 additions & 5 deletions handyrl/train.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -223,14 +223,13 @@ def compute_loss(batch, model, hidden, args):
outputs = map_r(outputs, lambda v: v[:, args['burn_in_steps']:])

actions = batch['action']
emasks = batch['episode_mask']
clip_rho_threshold, clip_c_threshold = 1.0, 1.0

log_selected_b_policies = torch.log(torch.clamp(batch['selected_prob'], 1e-16, 1)) * emasks
log_selected_t_policies = F.log_softmax(outputs['policy'], dim=-1).gather(-1, actions) * emasks
log_selected_b_policies = torch.log(torch.clamp(batch['selected_prob'], 1e-16, 1)) * batch['turn_mask']
log_selected_t_policies = F.log_softmax(outputs['policy'], dim=-1).gather(-1, actions) * batch['turn_mask']

# thresholds of importance sampling
log_rhos = log_selected_t_policies.detach() - log_selected_b_policies
log_rhos = (log_selected_t_policies.detach() - log_selected_b_policies).sum(-2, keepdim=True)
rhos = torch.exp(log_rhos)
clipped_rhos = torch.clamp(rhos, 0, clip_rho_threshold)
cs = torch.clamp(rhos, 0, clip_c_threshold)
Expand All @@ -241,7 +240,7 @@ def compute_loss(batch, model, hidden, args):
if args['turn_based_training'] and values_nograd.size(2) == 2: # two player zerosum game
values_nograd_opponent = -torch.stack([values_nograd[:, :, 1], values_nograd[:, :, 0]], dim=2)
values_nograd = (values_nograd + values_nograd_opponent) / (batch['observation_mask'].sum(dim=2, keepdim=True) + 1e-8)
outputs_nograd['value'] = values_nograd * emasks + batch['outcome'] * (1 - emasks)
outputs_nograd['value'] = values_nograd * batch['episode_mask'] + batch['outcome'] * (1 - batch['episode_mask'])

# compute targets and advantage
targets = {}
Expand Down