Fix bounded exploration

UoA-CARES · Jan 1, 2025 · 971884c · 971884c
1 parent e3cf728
commit 971884c
Show file tree

Hide file tree

Showing 2 changed files with 2 additions and 2 deletions.
diff --git a/cares_reinforcement_learning/algorithm/mbrl/DynaSAC_Bounded.py b/cares_reinforcement_learning/algorithm/mbrl/DynaSAC_Bounded.py
@@ -141,7 +141,7 @@ def select_action_from_policy(
                         multi_log_pi = multi_log_pi.squeeze()
                         policy_dist = F.softmax(multi_log_pi, dim=0)
 
-                        final_dist = policy_dist + self.threshold * world_dist
+                        final_dist = (1- self.threshold) * policy_dist + self.threshold * world_dist
                         final_dist = F.softmax(final_dist, dim=0)
                         # candi = torch.argmax(final_dist)
                         new_dist = torch.distributions.Categorical(final_dist)

diff --git a/cares_reinforcement_learning/algorithm/mbrl/STEVESAC_Bounded.py b/cares_reinforcement_learning/algorithm/mbrl/STEVESAC_Bounded.py
@@ -132,7 +132,7 @@ def select_action_from_policy(
 
                         multi_log_pi = multi_log_pi.squeeze()
                         policy_dist = F.softmax(multi_log_pi, dim=0)
-                        final_dist = policy_dist + self.threshold * world_dist
+                        final_dist = (1 - self.threshold) * policy_dist + self.threshold * world_dist
 
                         # candi = torch.argmax(final_dist)
                         final_dist = F.softmax(final_dist, dim=0)