Added DropConnect wrapper

keras_contrib/wrappers/__init__.py

@@ -0,0 +1 @@
+from .dropconnect import DropConnect

keras_contrib/wrappers/dropconnect.py

@@ -0,0 +1,52 @@
+from keras import backend as K
+from keras.layers.wrappers import Wrapper
+
+
+class DropConnect(Wrapper):
+    """
+    An implementation of DropConnect wrapper in Keras.
+    This layer drops connections between a one layer and
+    the next layer randomly with a given probability (rather
+    than dropping activations as in classic Dropout).
+
+    This wrapper can be used to drop the connections from
+    any Keras layer with weights and biases (Dense, LSTM etc)
+
+    #Example usage
+        dense = DropConnect(Dense(10, activation='sigmoid'), prob=0.05)
+        lstm = DropConnect(LSTM(20, activation='relu'), prob=0.2)
+
+    #Arguments
+        layer : Any Keras layer (instance of Layer class)
+        prob : dropout rate (probability)
+
+    #References
+        https://github.com/andry9454/KerasDropconnect/blob/master/ddrop/layers.py
+    """
+    def __init__(self, layer, prob=0.1, **kwargs):
+        self.prob = prob
+        self.layer = layer
+        super(DropConnect, self).__init__(layer, **kwargs)
+        if 0. < self.prob < 1.:
+            self.uses_learning_phase = True
+
+    def build(self, input_shape):
+        if not self.layer.built:
+            self.layer.build(input_shape)
+            self.layer.built = True
+
+    def compute_output_shape(self, input_shape):
+        return self.layer.compute_output_shape(input_shape)
+
+    def call(self, x):
+        if 0. < self.prob < 1.:
+            self.layer.kernel = K.in_train_phase(K.dropout(self.layer.kernel,
+                                                           self.prob),
+                                                 self.layer.kernel)
+            self.layer.bias = K.in_train_phase(K.dropout(self.layer.bias,
+                                                         self.prob),
+                                               self.layer.bias)
+        return self.layer.call(x)
+
+    def get_config(self):
+        return self.layer.get_config(self)