gym-examples 3.0.724__py3-none-any.whl → 3.0.726__py3-none-any.whl

gym_examples/__init__.py

@@ -5,4 +5,4 @@ register(
      entry_point="gym_examples.envs:WSNRoutingEnv",
 )
 
-__version__ = "3.0.724"
+__version__ = "3.0.726"

gym_examples/envs/wsn_env.py

@@ -211,7 +211,7 @@ class WSNRoutingEnv(gym.Env):
                 
             self.number_of_packets[i] = 0 # Reset the number of packets of the sensor i
             # Calculate final reward
-            rewards[i] = self.compute_attention_rewards(rewards[i])
+            # rewards[i] = self.compute_attention_rewards(rewards[i])
             # rewards[i] = np.mean(rewards[i])
         
         # Integrate the mobility of the sensors
@@ -225,8 +225,9 @@ class WSNRoutingEnv(gym.Env):
         self.get_metrics()
 
         rewards = [reward.item() if isinstance(reward, torch.Tensor) else reward for reward in rewards] # Convert the reward to a float
-        rewards = np.mean(rewards) # Average the rewards
-        # rewards = np.sum(rewards) # Sum the rewards
+        
+        rewards = self.compute_attention_reward(rewards)
+        rewards = np.sum(rewards)
 
         for i in range(self.n_sensors):
             if not dones[i]:
@@ -465,13 +466,44 @@ class WSNRoutingEnv(gym.Env):
         return np.clip(normalized_throughput, 0, 1)
     
 
-    def compute_attention_rewards(self, rewards):
+    # def compute_attention_rewards(self, rewards):
+    #     '''
+    #     Compute the attention-based rewards
+    #     '''
+    #     rewards = torch.tensor(rewards, dtype=torch.double)
+    #     final_reward = net(rewards)
+    #     return final_reward 
+
+
+    def compute_attention_reward(self, rewards):
         '''
-        Compute the attention-based rewards
+        Compute the attention-based reward for the network with Q, V, and K matrices
+
+        Input:
+                - rewards: list of rewards for each sensor
+        Output:
+                - final_reward: final reward for the network
         '''
-        rewards = torch.tensor(rewards, dtype=torch.double)
-        final_reward = net(rewards)
-        return final_reward 
+        # dimension of the sensor embedding
+        d = len(rewards[0])
+
+        # All sensors are represented by their raw rewards
+        query_vector = [max_reward] * d) # Basically, the target is the base station
+
+        # Similarities between the query vector and the rewards
+        similarities = [np.dot(query_vector, reward) for reward in rewards]
+
+        # Similarities scaling
+        similarities = [similarity / np.sqrt(d) for similarity in similarities]
+
+        # Softmax operation
+        denominator = sum([np.exp(similarity) for similarity in similarities])
+        attention_weights = [np.exp(similarity) / denominator for similarity in similarities]
+
+        # Weighted sum of the rewards
+        final_reward = sum([attention_weight * reward for attention_weight, reward in zip(attention_weights, rewards)])
+
+        return final_reward
 
 
     def integrate_mobility(self):

{gym_examples-3.0.724.dist-info → gym_examples-3.0.726.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: gym-examples
-Version: 3.0.724
+Version: 3.0.726
 Summary: A custom environment for multi-agent reinforcement learning focused on WSN routing.
 Home-page: https://github.com/gedji/CODES.git
 Author: Georges Djimefo

gym_examples-3.0.726.dist-info/RECORD

@@ -0,0 +1,7 @@
+gym_examples/__init__.py,sha256=uumjAP8o6H1zTcsXxnS9vkZq4QJsG507NyrsYL47Ffk,166
+gym_examples/envs/__init__.py,sha256=lgMe4pyOuUTgTBUddM0iwMlETsYTwFShny6ifm8PGM8,53
+gym_examples/envs/wsn_env.py,sha256=Z5OS65J_qwBorJdbeGBBN56xp4hey-IQZtvzH3dGxJo,26686
+gym_examples-3.0.726.dist-info/METADATA,sha256=vzWTlxeSCnU5TQSmoH1C4YxnBAJ5TPqplNLpHilLiLM,412
+gym_examples-3.0.726.dist-info/WHEEL,sha256=2wepM1nk4DS4eFpYrW1TTqPcoGNfHhhO_i5m4cOimbo,92
+gym_examples-3.0.726.dist-info/top_level.txt,sha256=rJRksoAF32M6lTLBEwYzRdo4PgtejceaNnnZ3HeY_Rk,13
+gym_examples-3.0.726.dist-info/RECORD,,

gym_examples-3.0.724.dist-info/RECORD

@@ -1,7 +0,0 @@
-gym_examples/__init__.py,sha256=Lm7GIk_8Yp9w3d60B2-sbWmMPMUX5vzsZl_kZoa7Lzw,166
-gym_examples/envs/__init__.py,sha256=lgMe4pyOuUTgTBUddM0iwMlETsYTwFShny6ifm8PGM8,53
-gym_examples/envs/wsn_env.py,sha256=D3EaMLo38IBUfq0CCGX8kmDnwwBtyKTEaNv4iIzxyn4,25477
-gym_examples-3.0.724.dist-info/METADATA,sha256=XPoX8NDFt5ZtoD8r8-7zuBANMVP5w3S3FIndfnJbhZs,412
-gym_examples-3.0.724.dist-info/WHEEL,sha256=2wepM1nk4DS4eFpYrW1TTqPcoGNfHhhO_i5m4cOimbo,92
-gym_examples-3.0.724.dist-info/top_level.txt,sha256=rJRksoAF32M6lTLBEwYzRdo4PgtejceaNnnZ3HeY_Rk,13
-gym_examples-3.0.724.dist-info/RECORD,,