Fixed weight name parser for depthwise-separable convolutions. Fixed activity plots for 1D convolutions.

Author: rbodo

snntoolbox/parsing/utils.py

@@ -393,7 +393,8 @@ def layers_to_skip(self):
                 'Activation',
                 'Dropout',
                 'ReLU',
-                'ActivityRegularization']
+                'ActivityRegularization',
+                'GaussianNoise']
 
     @abstractmethod
     def has_weights(self, layer):

snntoolbox/simulation/plotting.py

@@ -1154,8 +1154,8 @@ def plot_input_image(x, label, path=None, data_format=None, filename=None):
     if data_format == 'channels_first':
         x = np.transpose(x, (1, 2, 0))
 
-    if x.shape[-1] == 1 and x.ndim > 2:
-        x = x[..., 0]
+    if x.ndim > 2:
+        x = np.squeeze(x)
 
     plt.figure()
     plt.title('Input image (class: {})'.format(label))

snntoolbox/simulation/target_simulators/INI_temporal_mean_rate_target_sim.py

@@ -8,7 +8,6 @@
 from __future__ import print_function, unicode_literals
 
 import os
-import re
 import sys
 
 import keras
@@ -114,11 +113,22 @@ def compile(self):
         # Need to extract the actual weights here.
 
         def remove_name_counter(name_in):
-            splits = str(name_in).split('_')
-            name_out = splits[0] + '_' + splits[1]
-            if len(splits) == 3:
-                name_out += re.sub(r'\d+/', '/', splits[2])
-            return name_out
+            """
+            Tensorflow adds a counter to layer names, e.g. <name>/kernel:0 ->
+            <name>_0/kernel:0. Need to remove this _0.
+            Situation get complicated because SNN toolbox assigns layer names
+            that contain the layer shape, e.g. 00Conv2D_3x32x32. In addition,
+            we may get another underscore in the parameter name, e.g.
+            00DepthwiseConv2D_3X32x32_0/depthwise_kernel:0.
+            """
+
+            split_dash = str(name_in).split('/')
+            assert len(split_dash) == 2, "Layer name must not contain '/'."
+            # We are only interested in the part before the /.
+            split_underscore = split_dash[0].split('_')
+            # The first '_' is assigned by SNN toolbox and should be kept.
+            return (split_underscore[0] + '_' + split_underscore[1] + '/' +
+                    split_dash[1])
 
         parameter_map = {remove_name_counter(p.name): v for p, v in
                          zip(self.parsed_model.weights,