gym-examples 3.0.31__py3-none-any.whl → 3.0.32__py3-none-any.whl

gym_examples/__init__.py

@@ -6,4 +6,4 @@ register(
      max_episode_steps=50,
 )
 
-__version__ = "3.0.31"
+__version__ = "3.0.32"

gym_examples/envs/wsn_env.py

@@ -10,7 +10,7 @@ import torch.nn as nn
 import torch.nn.functional as F
 
 # Define the network parameters for the final reward function
-input_dim = 4  # number of individual rewards
+input_dim = 4  # lenght of the individual rewards vector
 output_dim = 1  # final reward
 
 Eelec = 50e-9  # energy consumption per bit in joules
@@ -40,11 +40,6 @@ class Attention(nn.Module):
         x = self.linear2(x)  # Pass the result through another linear layer
         return x
 
-# Calculate the reward
-net = Attention(input_dim, output_dim)
-net = net.double()  # Convert the weights to Double
-
-
 class WSNRoutingEnv(gym.Env):
     def __init__(self, n_sensors = 20, coverage_radius=(upper_bound - lower_bound)/4):
 
@@ -56,7 +51,6 @@ class WSNRoutingEnv(gym.Env):
         self.episode_count = 0
         self.scale_displacement = 0.01 * (upper_bound - lower_bound) # scale of the random displacement of the sensors
         self.epsilon = 1e-10 # small value to avoid division by zero
-        # self.rewards_individual = [0] * self.n_sensors
         # Initialize the position of the sensors randomly
 
         # Define observation space
@@ -129,7 +123,6 @@ class WSNRoutingEnv(gym.Env):
                 self.total_latency += self.packet_latency[i] + latency_per_hop
                 self.packet_latency[i] = 0
 
-                # rewards[i] = self.compute_individual_rewards(i, action)
                 rewards[i] = np.ones(input_dim) # input_dim should be equal to the number of individual rewards
                 dones[i] = True       
             else:
@@ -157,13 +150,7 @@ 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
             # Calculate final reward
-            rewards_individual = torch.tensor(rewards[i], dtype=torch.double)                
-            final_reward = net(rewards_individual)
-            # final_reward = np.sum(rewards[i]) 
-            # weights = np.ones(self.n_sensors, dtype=int)
-            # final_reward = np.sum(reward * weight for reward, weight in zip(rewards[i], weights))
-            rewards[i] = final_reward
-        # rewards = np.mean(rewards)
+            rewards[i] = self.compute_attention_rewards(rewards[i])
         for i in range(self.n_sensors):
             if (self.remaining_energy[i] <= 0) or (self.number_of_packets[i] <= 0):
                 dones[i] = True
@@ -178,12 +165,16 @@ class WSNRoutingEnv(gym.Env):
 
         self.get_metrics()
 
+        reward_packet_delivery_ratio = self.compute_reward_packet_delivery_ratio()
+        reward_latency = self.compute_reward_latency()
+        rewards_metrics = [reward_packet_delivery_ratio, reward_latency]
+        rewards_metrics = self.compute_attention_rewards(rewards_metrics)
+        rewards = np.array([r + rewards_metrics for r in rewards])
+
         return self._get_obs(), rewards, dones, {}
 
     def _get_obs(self):
 
-        performance = self.get_metrics()
-
         return [{'remaining_energy': np.array([e]), 
                  'consumption_energy': np.array([initial_energy - e]),
                  'sensor_positions': p,
@@ -301,7 +292,7 @@ class WSNRoutingEnv(gym.Env):
         Compute the individual rewards
         '''
         reward_angle = self.compute_reward_angle(i, action)
-        reward_distance = self.compute_reward_distance(i, action)
+        # reward_distance = self.compute_reward_distance(i, action)
         reward_consumption_energy = self.compute_reward_consumption_energy(i, action)
         reward_dispersion_remaining_energy = self.compute_reward_dispersion_remaining_energy()
         reward_number_of_packets = self.compute_reward_number_of_packets(action)
@@ -329,6 +320,39 @@ class WSNRoutingEnv(gym.Env):
         normalized_total_energy = total_energy / max_total_energy
 
         return np.clip(1 - normalized_total_energy, 0, 1)
+    
+    def compute_reward_packet_delivery_ratio(self):
+        '''
+        Compute the reward based on the packet delivery ratio
+        '''
+        return np.clip(self.packet_delivery_ratio, 0, 1)
+    
+    def compute_reward_latency(self):
+        '''
+        Compute the reward based on the average latency
+        '''
+        # Normalize the average latency
+        max_latency = self.n_sensors * self.steps
+        normalized_latency = self.total_latency / max_latency
+
+        return np.clip(1 - normalized_latency, 0, 1)
+
+
+    def compute_sum_rewards(self, rewards):
+        '''
+        Compute the sum of the rewards
+        '''
+        return np.sum(rewards)
+    
+    def compute_attention_rewards(self, rewards):
+        '''
+        Compute the attention-based rewards
+        '''
+        rewards = torch.tensor(rewards, dtype=torch.double)
+        net = Attention(len(rewards), output_dim)
+        net = net.double()  # Convert the weights to Double
+        final_reward = net(rewards)
+        return final_reward 
 
     def integrate_mobility(self):
         '''

{gym_examples-3.0.31.dist-info → gym_examples-3.0.32.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: gym-examples
-Version: 3.0.31
+Version: 3.0.32
 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-3.0.32.dist-info/RECORD

@@ -0,0 +1,7 @@
+gym_examples/__init__.py,sha256=tfSANHUypx5RPlDg703z8BuizUUYBVQT5JGSL3_a1Ec,193
+gym_examples/envs/__init__.py,sha256=lgMe4pyOuUTgTBUddM0iwMlETsYTwFShny6ifm8PGM8,53
+gym_examples/envs/wsn_env.py,sha256=G6SiehYtP1gWLyHeskzzEJGnI-QRnUCu8B0__gIOFIk,18900
+gym_examples-3.0.32.dist-info/METADATA,sha256=ZvRIjNWc1l-ElSzEC34USNseXM_-Rtcn-m4UIDaymnE,411
+gym_examples-3.0.32.dist-info/WHEEL,sha256=2wepM1nk4DS4eFpYrW1TTqPcoGNfHhhO_i5m4cOimbo,92
+gym_examples-3.0.32.dist-info/top_level.txt,sha256=rJRksoAF32M6lTLBEwYzRdo4PgtejceaNnnZ3HeY_Rk,13
+gym_examples-3.0.32.dist-info/RECORD,,

gym_examples-3.0.31.dist-info/RECORD

@@ -1,7 +0,0 @@
-gym_examples/__init__.py,sha256=augD2S6JxvSYGCFGUz5j2KohuhHRrqFqekAH7LtUdZ4,193
-gym_examples/envs/__init__.py,sha256=lgMe4pyOuUTgTBUddM0iwMlETsYTwFShny6ifm8PGM8,53
-gym_examples/envs/wsn_env.py,sha256=HDLGNMvbhGuweGA0IcJkhw1VuzdSiMTN0Ip-6rdvXUQ,18132
-gym_examples-3.0.31.dist-info/METADATA,sha256=IQDb4-6MmoLEdaGk9c2paIu5VrOp31WLclbjjW9M2mM,411
-gym_examples-3.0.31.dist-info/WHEEL,sha256=2wepM1nk4DS4eFpYrW1TTqPcoGNfHhhO_i5m4cOimbo,92
-gym_examples-3.0.31.dist-info/top_level.txt,sha256=rJRksoAF32M6lTLBEwYzRdo4PgtejceaNnnZ3HeY_Rk,13
-gym_examples-3.0.31.dist-info/RECORD,,