partial update for section 1

hw3/src/critics.py

@@ -41,7 +41,7 @@ class ValueCritic(nn.Module):
 
         ############################
         # YOUR IMPLEMENTATION HERE #
-        
+        values=self.network(obs)
         ############################
 
         return values
@@ -55,7 +55,12 @@ class ValueCritic(nn.Module):
         loss = None
         ############################
         # YOUR IMPLEMENTATION HERE #
+        values = self.forward(obs)
+        loss = F.mse_loss(values, q_values)
 
+        self.optimizer.zero_grad()
+        loss.backward()
+        self.optimizer.step()
         ############################
 
         return {

hw3/src/pg_agent.py

@@ -148,10 +148,11 @@ class PGAgent(nn.Module):
             advantages = None
             ############################
             # YOUR IMPLEMENTATION HERE #
-            source = rewards.copy()
+            source = q_values.copy()
             mean = np.mean(source)
             std = np.std(source)
-            advantages = (source - mean)/std
+            # avoid division by zero
+            advantages = (source - mean)/(std+1e-8)
             ############################
 
         return advantages