gym-examples 3.0.376__py3-none-any.whl → 3.0.378__py3-none-any.whl

gym_examples/__init__.py

@@ -5,4 +5,4 @@ register(
      entry_point="gym_examples.envs:WSNRoutingEnv",
 )
 
-__version__ = "3.0.376"
+__version__ = "3.0.378"

gym_examples/envs/wsn_env.py

@@ -29,32 +29,31 @@ base_back_up_dir = "results/data/"
 max_reward = 1 # maximum reward value when the sensors sent data to the base station. The opposite value is when the sensors perform an unauthorized action
 
 # Define the final reward function using an attention mechanism
-class CustomizedLinear(nn.Module):
-    def __init__(self, input_dim, output_dim):
-        super(CustomizedLinear, self).__init__()
-        self.weight = nn.Parameter(torch.rand(output_dim, input_dim))
-        self.bias = None  # No bias term
-        self.Softplus = nn.Softplus()  # SoftPlus activation function to ensure non-negative values: Check the paper for more details
-
-    def forward(self, x):
-        # Enforce non-negativity of weights
-        weight = self.Softplus(self.weight)
+# class CustomizedLinear(nn.Module):
+#     def __init__(self, input_dim, output_dim):
+#         super(CustomizedLinear, self).__init__()
+#         self.weight = nn.Parameter(torch.rand(output_dim, input_dim))
+#         self.bias = None  # No bias term
         
-        # Normalize cols to ensure that if sum(x1) < sum(x2) ==> sum(Ax1 + 0) < sum(Ax2 + 0): proof in the paper
-        col_sums = weight.sum(dim=0, keepdim=True)
-        normalized_weight = weight / col_sums  
+#     def forward(self, x):        
+#         # Normalize cols to ensure that if sum(x1) < sum(x2) ==> sum(Ax1 + 0) < sum(Ax2 + 0): proof in the paper
+#         col_sums = self.weight.sum(dim=0, keepdim=True)
+#         normalized_weight = self.weight / col_sums  
         
-        # Output
-        y = torch.matmul(x, normalized_weight.t())
-        return y
+#         # Output
+#         y = torch.matmul(x, normalized_weight.t())
+#         return y
 
 class Attention(nn.Module):
     def __init__(self, input_dim, output_dim):
         super(Attention, self).__init__()  # Call the initializer of the parent class (nn.Module)
         self.input_dim = input_dim  # Set the input dimension of the network
         self.output_dim = output_dim  # Set the output dimension of the network
-        self.linear1 = CustomizedLinear(input_dim, 64)  # Define the first linear layer. It takes input of size 'input_dim' and outputs size '64'
-        self.linear2 = CustomizedLinear(64, output_dim)  # Define the second linear layer. It takes input of size '64' and outputs size 'output_dim'
+        # self.linear1 = CustomizedLinear(input_dim, 64)  # Define the first linear layer. It takes input of size 'input_dim' and outputs size '64'
+        # self.linear2 = CustomizedLinear(64, output_dim)  # Define the second linear layer. It takes input of size '64' and outputs size 'output_dim'
+        self.linear1 = nn.Linear(input_dim, 64)  # Define the first linear layer. It takes input of size 'input_dim' and outputs size '64'
+        self.linear2 = nn.Linear(64, output_dim)  # Define the second linear layer. It takes input of size '64' and outputs size 'output_dim'
+
 
     def forward(self, x):
         # Step 1: Ensure input is 2D by adding a batch dimension if necessary
@@ -197,6 +196,7 @@ class WSNRoutingEnv(gym.Env):
                 self.packet_latency[i] = 0
 
                 rewards[i] = [max_reward] * input_dim # Reward for transmitting data to the base station
+                print(f"Sensor {i} transmitted data to the base station with modified reward: {self.compute_attention_rewards(rewards[i])}")
                 dones[i] = True
             else:
                 distance = np.linalg.norm(self.sensor_positions[i] - self.sensor_positions[action])
@@ -222,7 +222,11 @@ class WSNRoutingEnv(gym.Env):
                 self.packet_latency[action] += self.packet_latency[i] + latency_per_hop
                 self.packet_latency[i] = 0
 
-                rewards[i] = self.compute_individual_rewards(i, action) 
+                rewards[i] = self.compute_individual_rewards(i, action)
+                if self.number_steps > 27:
+                    raise Error("Stop here")
+                else:
+                    print(f"Sensor {i} transmitted data to sensor {action} with modified reward: {self.compute_attention_rewards(rewards[i])}") 
 
                 # Update the number of packets
                 self.number_of_packets[action] += self.number_of_packets[i]

{gym_examples-3.0.376.dist-info → gym_examples-3.0.378.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: gym-examples
-Version: 3.0.376
+Version: 3.0.378
 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-3.0.378.dist-info/RECORD

@@ -0,0 +1,7 @@
+gym_examples/__init__.py,sha256=rkNQCsuNv0lbzwDFbbINy72Zb5H42sq2w5lvX-kVL5k,166
+gym_examples/envs/__init__.py,sha256=lgMe4pyOuUTgTBUddM0iwMlETsYTwFShny6ifm8PGM8,53
+gym_examples/envs/wsn_env.py,sha256=UITF_EWvfj15xD39FPBqig-tCSh2XSLHLUDcnKGATtU,27260
+gym_examples-3.0.378.dist-info/METADATA,sha256=23rCX5hkl1xinlpHNHZIdFH1-F99ZtUk03yRGEek-U0,412
+gym_examples-3.0.378.dist-info/WHEEL,sha256=2wepM1nk4DS4eFpYrW1TTqPcoGNfHhhO_i5m4cOimbo,92
+gym_examples-3.0.378.dist-info/top_level.txt,sha256=rJRksoAF32M6lTLBEwYzRdo4PgtejceaNnnZ3HeY_Rk,13
+gym_examples-3.0.378.dist-info/RECORD,,

gym_examples-3.0.376.dist-info/RECORD

@@ -1,7 +0,0 @@
-gym_examples/__init__.py,sha256=saXnZJjlgXXuJWFxaY2VD_0xdoYKK3gEsrD5TPNdybE,166
-gym_examples/envs/__init__.py,sha256=lgMe4pyOuUTgTBUddM0iwMlETsYTwFShny6ifm8PGM8,53
-gym_examples/envs/wsn_env.py,sha256=CDqmuxieq6URlTpDQWgDcOaKJVhhzd6h4jrh72WvGh4,26756
-gym_examples-3.0.376.dist-info/METADATA,sha256=ZIc51ev191-r6-0hK2-dntjvgjPinlgtJ2UAxbQHEnQ,412
-gym_examples-3.0.376.dist-info/WHEEL,sha256=2wepM1nk4DS4eFpYrW1TTqPcoGNfHhhO_i5m4cOimbo,92
-gym_examples-3.0.376.dist-info/top_level.txt,sha256=rJRksoAF32M6lTLBEwYzRdo4PgtejceaNnnZ3HeY_Rk,13
-gym_examples-3.0.376.dist-info/RECORD,,