flexible initialization of weights

neural_filters/__init__.py

+RANDOM_STD = 0
+
 from .neural_filter import *
 from .neural_filter_1L import *
 from .neural_filter_2R import *

neural_filters/neural_filter.py

@@ -30,6 +30,7 @@ from torch.nn import Parameter
 from torch.nn import functional as F
 import numpy as np
 
+from . import RANDOM_STD
 
 class NeuralFilter(torch.nn.Module):
     """
@@ -49,9 +50,13 @@ class NeuralFilter(torch.nn.Module):
 
     def reset_parameters(self, init=None):
         if init is None:
-            self.bias_forget.data.uniform_(-0.2, 0.2)
+            self.bias_forget.data.uniform_(-RANDOM_STD, RANDOM_STD)
         else:
-            self.bias_forget.data.fill_(init)
+            if not isinstance(init, np.ndarray):
+                init = np.array(init, ndmin=1)
+
+            ten_init = torch.from_numpy(init)
+            self.bias_forget.data.copy_(ten_init)
 
     def __repr__(self):
         s = '{name}({hidden_size})'

neural_filters/neural_filter_2CC.py

@@ -30,6 +30,8 @@ from torch.nn import Parameter
 from torch.nn import functional as F
 import numpy as np
 
+from . import RANDOM_STD
+
 
 class NeuralFilter2CC(torch.nn.Module):
     """
@@ -48,17 +50,26 @@ class NeuralFilter2CC(torch.nn.Module):
 
         self.reset_parameters()
 
-    def reset_parameters(self, init=None):
-        if init is None:
-            self.bias_modulus.data.uniform_(-0.2, 0.2)
-            self.bias_theta.data.uniform_(-0.2, 0.2)
+    def reset_parameters(self, init_modulus=None, init_theta=None):
+        if init_modulus is None:
+            self.bias_modulus.data.uniform_(-RANDOM_STD, RANDOM_STD)
+
         else:
-            if isinstance(init, tuple):
-                self.bias_modulus.data.fill_(init[0])
-                self.bias_theta.data.fill_(init[1])
+            if not isinstance(init_modulus, np.ndarray):
+                init_modulus = np.array(init_modulus, ndmin=1)
+
+            ten_init = torch.from_numpy(init_modulus)
+            self.bias_modulus.data.copy_(ten_init)
+
+        if init_theta is None:
+            self.bias_theta.data.uniform_(-RANDOM_STD, RANDOM_STD)
+
         else:
-                self.bias_theta.data.fill_(init)
-                self.bias_modulus.data.fill_(init)
+            if not isinstance(init_theta, np.ndarray):
+                init_theta = np.array(init_theta, ndmin=1)
+
+            ten_init = torch.from_numpy(init_theta)
+            self.bias_theta.data.copy_(ten_init)
 
     def __repr__(self):
         s = '{name}({hidden_size})'

neural_filters/neural_filter_2R.py

@@ -30,6 +30,8 @@ from . import NeuralFilter
 import torch
 import numpy as np
 
+from . import RANDOM_STD
+
 
 class NeuralFilter2R(torch.nn.Module):
     """
@@ -50,8 +52,8 @@ class NeuralFilter2R(torch.nn.Module):
 
     def reset_parameters(self, init=None):
         if init is None:
-            self.first_cell.bias_forget.data.uniform_(-0.7, -0.3)
-            self.second_cell.bias_forget.data.uniform_(0.3, 0.7)
+            self.first_cell.bias_forget.data.uniform_(-0.5 - RANDOM_STD, -0.5 + RANDOM_STD)
+            self.second_cell.bias_forget.data.uniform_(0.5 - RANDOM_STD, 0.5 + RANDOM_STD)
         elif isinstance(init, tuple):
             self.first_cell.reset_parameters(init[0])
             self.second_cell.reset_parameters(init[1])