gym-examples 3.0.402__py3-none-any.whl → 3.0.404__py3-none-any.whl

gym_examples/__init__.py

@@ -5,4 +5,4 @@ register(
      entry_point="gym_examples.envs:WSNRoutingEnv",
 )
 
-__version__ = "3.0.402"
+__version__ = "3.0.404"

gym_examples/envs/wsn_env.py

@@ -7,6 +7,7 @@ import csv
 from datetime import datetime
 import torch
 import torch.nn as nn
+from gaussian_adaptive_attention import GaussianAdaptiveAttention
 import torch.nn.functional as F
 import os
 from collections import OrderedDict
@@ -28,29 +29,51 @@ latency_per_hop = 1  # latency per hop in seconds
 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
 
-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 = 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'
-
+# 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 = 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
+#         if x.dim() == 1:
+#             x = x.unsqueeze(0)  # Shape: [1, input_dim]
+#         # x = F.relu(self.linear1(x))  # Pass the input through a linear layer and a ReLU activation function 
+#         x = F.softplus(self.linear1(x))  # Pass the input through a linear layer and a ReLU activation function       
+#         attention_weights = F.softmax(x, dim=1)  # Apply the softmax function to get the attention weights. Initially F.softmax(x, dim=0) 
+#         x = attention_weights * x  # Multiply the input by the attention weights
+#         x = self.linear2(x)  # Pass the result through another linear layer
+#         return x
+
+# net = Attention(input_dim, output_dim)
+# net = net.double()  # Convert the weights to Double
+
+class ScalarAttentionModel(nn.Module):
+    def __init__(self, input_dim):
+        super(ScalarAttentionModel, self).__init__()
+        # Initialize GaussianAdaptiveAttention
+        self.ga_attention = GaussianAdaptiveAttention(
+            num_heads=1,  # Single head for simplicity
+            num_gaussians=4,  # Corresponds to your input length
+            attention_axis=0,  # Attention along the first axis
+        )
+        self.output_layer = nn.Linear(input_dim, 1)  # Map to scalar output
 
     def forward(self, x):
-        # Step 1: Ensure input is 2D by adding a batch dimension if necessary
-        if x.dim() == 1:
-            x = x.unsqueeze(0)  # Shape: [1, input_dim]
-        # x = F.relu(self.linear1(x))  # Pass the input through a linear layer and a ReLU activation function 
-        x = F.softplus(self.linear1(x))  # Pass the input through a linear layer and a ReLU activation function       
-        attention_weights = F.softmax(x, dim=1)  # Apply the softmax function to get the attention weights. Initially F.softmax(x, dim=0) 
-        x = attention_weights * x  # Multiply the input by the attention weights
-        x = self.linear2(x)  # Pass the result through another linear layer
-        return x
-
-net = Attention(input_dim, output_dim)
+        # Apply GaussianAdaptiveAttention
+        attention_output = self.ga_attention(x.unsqueeze(0))  # Add batch dim
+        attention_output = attention_output.squeeze(0)  # Remove batch dim
+        # Map to scalar
+        scalar_output = self.output_layer(attention_output)
+        return scalar_output
+
+net = ScalarAttentionModel(input_dim)
 net = net.double()  # Convert the weights to Double
 
 class WSNRoutingEnv(gym.Env):

{gym_examples-3.0.402.dist-info → gym_examples-3.0.404.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: gym-examples
-Version: 3.0.402
+Version: 3.0.404
 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-3.0.404.dist-info/RECORD

@@ -0,0 +1,7 @@
+gym_examples/__init__.py,sha256=kSSUOydEpYd6p-FwjBNXhpeRAH5NAZ7lXf2MEHD66RE,166
+gym_examples/envs/__init__.py,sha256=lgMe4pyOuUTgTBUddM0iwMlETsYTwFShny6ifm8PGM8,53
+gym_examples/envs/wsn_env.py,sha256=hLM8UtyVrLn-dm_8ZazOjR_iuBk9mLxzCwRRsIbopWs,27341
+gym_examples-3.0.404.dist-info/METADATA,sha256=vrZnbnEa5pzhM2LTWiayHm5fMGebm-KKeVxMyBG_yuE,412
+gym_examples-3.0.404.dist-info/WHEEL,sha256=2wepM1nk4DS4eFpYrW1TTqPcoGNfHhhO_i5m4cOimbo,92
+gym_examples-3.0.404.dist-info/top_level.txt,sha256=rJRksoAF32M6lTLBEwYzRdo4PgtejceaNnnZ3HeY_Rk,13
+gym_examples-3.0.404.dist-info/RECORD,,

gym_examples-3.0.402.dist-info/RECORD

@@ -1,7 +0,0 @@
-gym_examples/__init__.py,sha256=QqlM4kYrw8groETIibxHM_PQxmIt-bj9-Na76C4w5Lg,166
-gym_examples/envs/__init__.py,sha256=lgMe4pyOuUTgTBUddM0iwMlETsYTwFShny6ifm8PGM8,53
-gym_examples/envs/wsn_env.py,sha256=4t_GZDrpm_6OJtMK2AE3mz5km5x5mX8yXagdJk2aMC4,26284
-gym_examples-3.0.402.dist-info/METADATA,sha256=bTRc0E0PkHNEIVus7x26qfqXkiV4MtI1t4BWi__87zs,412
-gym_examples-3.0.402.dist-info/WHEEL,sha256=2wepM1nk4DS4eFpYrW1TTqPcoGNfHhhO_i5m4cOimbo,92
-gym_examples-3.0.402.dist-info/top_level.txt,sha256=rJRksoAF32M6lTLBEwYzRdo4PgtejceaNnnZ3HeY_Rk,13
-gym_examples-3.0.402.dist-info/RECORD,,