gym-examples 3.0.762__py3-none-any.whl → 3.0.764__py3-none-any.whl

gym_examples/__init__.py

@@ -5,4 +5,4 @@ register(
      entry_point="gym_examples.envs:WSNRoutingEnv",
 )
 
-__version__ = "3.0.762"
+__version__ = "3.0.764"

gym_examples/envs/wsn_env.py

@@ -150,7 +150,7 @@ class WSNRoutingEnv(gym.Env):
         self.number_of_steps += 1
         self.steps += 1 
         # rewards = [-max_reward] * self.n_sensors
-        reward_init = [0] * input_dim
+        reward_init = np.array([0] * input_dim)
         rewards = [reward_init] * self.n_sensors
         dones = [False] * self.n_sensors
         for i, action in enumerate(actions):
@@ -179,7 +179,7 @@ class WSNRoutingEnv(gym.Env):
                 self.total_latency += self.packet_latency[i] + latency_per_hop
                 self.packet_latency[i] = 0
 
-                rewards[i] = [max_reward] * input_dim # Reward for transmitting data to the base station
+                rewards[i] = np.array([max_reward] * input_dim) # Reward for transmitting data to the base station
                 dones[i] = True
             else:
                 distance = np.linalg.norm(self.sensor_positions[i] - self.sensor_positions[action])
@@ -211,10 +211,6 @@ class WSNRoutingEnv(gym.Env):
                 self.number_of_packets[action] += self.number_of_packets[i]
                 
             self.number_of_packets[i] = 0 # Reset the number of packets of the sensor i
-
-        for i in range(self.n_sensors):
-            # Calculate final reward
-            rewards[i] = self.compute_attention_rewards(rewards[i])
         
         # Integrate the mobility of the sensors
         # self.integrate_mobility() 
@@ -227,7 +223,7 @@ 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)
+        rewards = self.compute_attention_reward(rewards) # Compute the attention-based reward
 
         for i in range(self.n_sensors):
             if not dones[i]:
@@ -381,7 +377,7 @@ class WSNRoutingEnv(gym.Env):
         reward_dispersion_remaining_energy = self.compute_reward_dispersion_remaining_energy()
         reward_number_of_packets = self.compute_reward_number_of_packets(action)
 
-        rewards_energy = [reward_angle, reward_consumption_energy, reward_dispersion_remaining_energy, reward_number_of_packets]
+        rewards_energy = np.array([reward_angle, reward_consumption_energy, reward_dispersion_remaining_energy, reward_number_of_packets])
 
         #-- rewards related to the performance metrics
         reward_latency = self.compute_reward_latency()
@@ -389,7 +385,7 @@ class WSNRoutingEnv(gym.Env):
         reward_network_throughput = self.compute_reward_network_throughput()
         reward_packet_delivery_ratio = self.compute_reward_packet_delivery_ratio()
 
-        rewards_performance = [reward_latency, reward_network_throughput, reward_packet_delivery_ratio]
+        rewards_performance = np.array([reward_latency, reward_network_throughput, reward_packet_delivery_ratio])
     
         return rewards_energy
     
@@ -464,45 +460,44 @@ class WSNRoutingEnv(gym.Env):
         return np.clip(normalized_throughput, 0, 1)
     
 
-    def compute_attention_rewards(self, reward):
-        '''
-        Compute the attention-based rewards
-        '''
-        rewards_i = torch.tensor(reward, dtype=torch.double)
-        rewards_i = rewards_i.unsqueeze(0) # Add batch dimension
-        print(f"net(rewards_i).item(): {net(rewards_i).item()}")
-        return np.mean(net(rewards_i).item()) 
+    # def compute_attention_rewards(self, reward):
+    #     '''
+    #     Compute the attention-based rewards
+    #     '''
+    #     rewards_i = torch.tensor(reward, dtype=torch.double)
+    #     rewards_i = rewards_i.unsqueeze(0) # Add batch dimension
+    #     return net(rewards_i).item()
 
 
-    # def compute_attention_reward(self, rewards):
-    #     '''
-    #     Compute the attention-based reward for the network with Q, V, and K matrices
+    def compute_attention_reward(self, 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
-    #     '''
-    #     # dimension of the sensor embedding
-    #     d = len(rewards[0])
+        Input:
+                - rewards: list of rewards for each sensor
+        Output:
+                - final_reward: final reward for the network
+        '''
+        # dimension of the sensor embedding
+        d = len(rewards[0])
 
-    #     # All sensors are represented by their raw rewards
-    #     query_vector = np.array([max_reward] * d) # Basically, the target is the base station
+        # All sensors are represented by their raw rewards
+        query_vector = np.array([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 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]
+        # 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]
+        # 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)])
+        # Weighted sum of the rewards
+        final_reward = sum([attention_weight * reward for attention_weight, reward in zip(attention_weights, rewards)])
 
-    #     return final_reward
+        return final_reward
 
 
     def integrate_mobility(self):

{gym_examples-3.0.762.dist-info → gym_examples-3.0.764.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: gym-examples
-Version: 3.0.762
+Version: 3.0.764
 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-3.0.764.dist-info/RECORD

@@ -0,0 +1,7 @@
+gym_examples/__init__.py,sha256=iGclI8newmGbCQqTxcbiIdqdKpVF-L4HIE49f_Ad9qQ,166
+gym_examples/envs/__init__.py,sha256=lgMe4pyOuUTgTBUddM0iwMlETsYTwFShny6ifm8PGM8,53
+gym_examples/envs/wsn_env.py,sha256=mlurW0_198GO2zyTX2xlU5KrrodDUoES1pxPDGWAq6k,26452
+gym_examples-3.0.764.dist-info/METADATA,sha256=OYFWn9IOfYO_l0MHyDiedJLEO8T6ocdiBBLS9OLmdOY,412
+gym_examples-3.0.764.dist-info/WHEEL,sha256=2wepM1nk4DS4eFpYrW1TTqPcoGNfHhhO_i5m4cOimbo,92
+gym_examples-3.0.764.dist-info/top_level.txt,sha256=rJRksoAF32M6lTLBEwYzRdo4PgtejceaNnnZ3HeY_Rk,13
+gym_examples-3.0.764.dist-info/RECORD,,

gym_examples-3.0.762.dist-info/RECORD

@@ -1,7 +0,0 @@
-gym_examples/__init__.py,sha256=fQGHHtmkYG0B6YCGmVYN9lXXBt2FkWSOzbzbxeok_-Q,166
-gym_examples/envs/__init__.py,sha256=lgMe4pyOuUTgTBUddM0iwMlETsYTwFShny6ifm8PGM8,53
-gym_examples/envs/wsn_env.py,sha256=DpQTrUKg2pwCFgRos3uFzMSBEzkarkPxMPVwUIufOUg,26609
-gym_examples-3.0.762.dist-info/METADATA,sha256=NwHlvSZZu_HmvnJZ8Vusc8MseOysNnaJAjZ1VWyuN_g,412
-gym_examples-3.0.762.dist-info/WHEEL,sha256=2wepM1nk4DS4eFpYrW1TTqPcoGNfHhhO_i5m4cOimbo,92
-gym_examples-3.0.762.dist-info/top_level.txt,sha256=rJRksoAF32M6lTLBEwYzRdo4PgtejceaNnnZ3HeY_Rk,13
-gym_examples-3.0.762.dist-info/RECORD,,