Increase number of channels

configs/train.json

@@ -13,6 +13,7 @@
     "conv_dropout": 0.25,
     "num_skip": 2,
     "is_correct_resnet": false,
+    "conv_num_filters_start": 32,
 
     "learning_rate": 0.001,
     "optimizer": "adam",

ecg/network.py

@@ -23,28 +23,54 @@ def _bn_relu(layer, **params):
     return layer
 
 
-def add_conv_weight(layer, filter_length, subsample_length, **params):
+def add_conv_weight(
+        layer,
+        filter_length,
+        num_filters,
+        subsample_length=1,
+        **params):
     from keras.layers.convolutional import Convolution1D
     layer = Convolution1D(
-        nb_filter=params["conv_num_filters"],
+        nb_filter=num_filters,
         filter_length=filter_length,
         border_mode='same',
         subsample_length=subsample_length,
         init=params["conv_init"])(layer)
     return layer
 
 
-def resnet_block(layer, subsample_length, **params):
+def resnet_block(
+        layer,
+        num_filters,
+        subsample_length,
+        zero_pad=False,
+        **params):
     from keras.layers import merge
     from keras.layers.pooling import MaxPooling1D
 
+    from keras.layers.core import Lambda
+    from keras import backend as K
+
+    def zeropad(x):
+        y = K.zeros_like(x)
+        return K.concatenate([x, y], axis=2)
+
+    def zeropad_output_shape(input_shape):
+        shape = list(input_shape)
+        assert len(shape) == 3
+        shape[2] *= 2
+        return tuple(shape)
+
     shortcut = MaxPooling1D(pool_length=subsample_length)(layer)
+    if zero_pad is True:
+        shortcut = Lambda(zeropad, output_shape=zeropad_output_shape)(shortcut)
 
     for i in range(params["num_skip"]):
         layer = _bn_relu(layer, **params)
         layer = add_conv_weight(
             layer,
             params["conv_filter_length"],
+            num_filters,
             subsample_length if i == 0 else 1,
             **params)
 
@@ -53,19 +79,33 @@ def resnet_block(layer, subsample_length, **params):
 
 
 def add_resnet_layers(layer, **params):
-    layer = _bn_relu(add_conv_weight(layer, 16, 1, **params), **params)
-    for subsample_length in params["conv_subsample_lengths"]:
-        layer = resnet_block(layer, subsample_length, **params)
+    layer = add_conv_weight(
+        layer,
+        params["conv_filter_length"],
+        params["conv_num_filters_start"],
+        subsample_length=1,
+        **params)
+    layer = _bn_relu(layer, **params)
+    for index, subsample_length in enumerate(params["conv_subsample_lengths"]):
+        num_filters = 2**int(index / 2) * params["conv_num_filters_start"]
+        zero_pad = (index % 2) == 0 and index > 0
+        layer = resnet_block(
+            layer,
+            num_filters,
+            subsample_length,
+            zero_pad=zero_pad,
+            **params)
     layer = _bn_relu(layer, **params)
     return layer
 
 
 def add_conv_layers(layer, **params):
     from keras.layers import merge
-    for subsample_length in params["conv_subsample_lengths"]:
+    for index, subsample_length in enumerate(params["conv_subsample_lengths"]):
         shortcut = add_conv_weight(
             layer,
             params["conv_filter_length"],
+            params["conv_num_filters"],
             subsample_length,
             **params)
         layer = shortcut
@@ -74,7 +114,8 @@ def add_conv_layers(layer, **params):
             layer = add_conv_weight(
                 layer,
                 params["conv_filter_length"],
-                1,
+                params["conv_num_filters"],
+                subsample_length=1,
                 **params)
         layer = merge([shortcut, layer], mode="sum")
     return layer