fix unroll of BidirectionalCell

python/mxnet/gluon/rnn/rnn_cell.py

@@ -102,21 +102,21 @@ def _mask_sequence_variable_length(F, data, length, valid_length, time_axis, mer
                                    squeeze_axis=True))
     return outputs
 
-def _reverse_sequence(sequences, time_step, valid_length=None):
+def _reverse_sequences(sequences, unroll_step, valid_length=None):
     if isinstance(sequences[0], symbol.Symbol):
         F = symbol
     else:
         F = ndarray
 
     if valid_length is None:
-        reversed_inputs = list(reversed(sequences))
+        reversed_sequences = list(reversed(sequences))
     else:
-        reversed_inputs = F.SequenceReverse(F.stack(*sequences, axis=0),
-                                            sequence_length=valid_length,
-                                            use_sequence_length=True)
-        reversed_inputs = F.split(reversed_inputs, axis=0, num_outputs=time_step, squeeze_axis=True)
+        reversed_sequences = F.SequenceReverse(F.stack(*sequences, axis=0),
+                                               sequence_length=valid_length,
+                                               use_sequence_length=True)
+        reversed_sequences = F.split(reversed_sequences, axis=0, num_outputs=unroll_step, squeeze_axis=True)
 
-    return reversed_inputs
+    return reversed_sequences
 
 
 class RecurrentCell(Block):
@@ -1052,7 +1052,8 @@ def unroll(self, length, inputs, begin_state=None, layout='NTC', merge_outputs=N
         self.reset()
 
         inputs, axis, F, batch_size = _format_sequence(length, inputs, layout, False)
-        reversed_inputs = list(_reverse_sequence(inputs, length, valid_length))
+        reversed_inputs = list(_reverse_sequences(inputs, length, valid_length))
+        begin_state = _get_begin_state(self, F, begin_state, inputs, batch_size)
 
         states = begin_state
         l_cell, r_cell = self._children.values()
@@ -1065,7 +1066,7 @@ def unroll(self, length, inputs, begin_state=None, layout='NTC', merge_outputs=N
                                             begin_state=states[len(l_cell.state_info(batch_size)):],
                                             layout=layout, merge_outputs=False,
                                             valid_length=valid_length)
-        reversed_r_outputs = _reverse_sequence(r_outputs, length, valid_length)
+        reversed_r_outputs = _reverse_sequences(r_outputs, length, valid_length)
 
         if merge_outputs is None:
             merge_outputs = isinstance(l_outputs, tensor_types)