Fix merge conflicts

keras/activations.py

@@ -7,13 +7,9 @@ def softmax(x):
     if ndim == 2:
         return K.softmax(x)
     elif ndim == 3:
-        # apply softmax to each timestep
-        def step(x, states):
-            return K.softmax(x), []
-        last_output, outputs, states = K.rnn(step, x,
-                                             [],
-                                             mask=None)
-        return outputs
+        e = K.exp(x - K.max(x, axis=-1, keepdims=True))
+        s = K.sum(e, axis=-1, keepdims=True)
+        return e / s
     else:
         raise Exception('Cannot apply softmax to a tensor that is not 2D or 3D. ' +
                         'Here, ndim=' + str(ndim))

keras/backend/tensorflow_backend.py

@@ -353,6 +353,8 @@ def spatial_2d_padding(x, padding=(1, 1), dim_ordering='th'):
                    [0, 0]]
     return tf.pad(x, pattern)
 
+def pack(x):
+    return tf.pack(x)
 
 # VALUE MANIPULATION
 

keras/backend/theano_backend.py

@@ -415,6 +415,8 @@ def spatial_3d_padding(x, padding=(1, 1, 1), dim_ordering='th'):
         raise Exception('Invalid dim_ordering: ' + dim_ordering)
     return T.set_subtensor(output[indices], x)
 
+def pack(x):
+    return T.stack(*x)
 
 # VALUE MANIPULATION
 

keras/layers/core.py

@@ -1145,17 +1145,18 @@ class TimeDistributedDense(MaskedLayer):
         return (input_shape[0], input_shape[1], self.output_dim)
 
     def get_output(self, train=False):
-        X = self.get_input(train)
-
-        def step(x, states):
-            output = K.dot(x, self.W) + self.b
-            return output, []
-
-        last_output, outputs, states = K.rnn(step, X,
-                                             initial_states=[],
-                                             mask=None)
-        outputs = self.activation(outputs)
-        return outputs
+        X = self.get_input(train)  # (samples, timesteps, input_dim)
+        # Squash samples and timesteps into a single axis
+        x = K.reshape(X, (-1, self.input_shape[-1]))  # (samples * timesteps, input_dim)
+        Y = K.dot(x, self.W) + self.b  # (samples * timesteps, output_dim)
+        # We have to reshape Y to (samples, timesteps, output_dim)
+        input_length = self.input_shape[1]
+        # Note: input_length will always be provided when using tensorflow backend.
+        if not input_length:
+            input_length = K.shape(X)[1]
+        Y = K.reshape(Y, (-1, input_length, self.output_shape[-1]))  # (samples, timesteps, output_dim)
+        Y = self.activation(Y)
+        return Y
 
     def get_config(self):
         config = {'name': self.__class__.__name__,

keras/layers/recurrent.py

@@ -149,10 +149,9 @@ class Recurrent(MaskedLayer):
 
     def get_initial_states(self, x):
         # build an all-zero tensor of shape (samples, output_dim)
-        initial_state = K.zeros_like(x)  # (samples, timesteps, input_dim)
-        initial_state = K.sum(initial_state, axis=1)  # (samples, input_dim)
-        reducer = K.zeros((self.input_dim, self.output_dim))
-        initial_state = K.dot(initial_state, reducer)  # (samples, output_dim)
+        initial_state = x[:, 0, 0] * 0  # (samples, )
+        initial_state = K.pack([initial_state] * self.output_dim)  # (output_dim, samples)
+        initial_state = K.permute_dimensions(initial_state, (1, 0))  # (samples, output_dim)
         initial_states = [initial_state for _ in range(len(self.states))]
         return initial_states