Hotfix tweak convolution

xnmt/specialized_encoders/segmenting_encoder/segmenting_composer.py

@@ -42,11 +42,15 @@ def compose(self, composed_words, sample_size, batch_size):
       now_idx += 1
     # Composing
     outputs = [[[None for _ in range(seq_len[i,j])] for j in range(batch_size)] for i in range(sample_size)]
+    expr_list = []
     for batch, batch_map, batch_word in zip(batches, batch_maps, batch_words):
       self.set_words(batch_word)
       results = self.transduce(dy.concatenate_to_batch(batch))
+      results.value()
       for idx, (sample_num, batch_num, position) in batch_map.items():
-        outputs[sample_num][batch_num][position] = dy.pick_batch_elem(results, idx)
+        expr_list.append(dy.pick_batch_elem(results, idx))
+        outputs[sample_num][batch_num][position] = expr_list[-1]
+    dy.forward(expr_list)
     return outputs
 
   @handle_xnmt_event
@@ -125,7 +129,7 @@ def transduce(self, encodings):
       pad = dy.zeros((self.embed_dim, self.ngram_size-dim[0][1]))
       inp = dy.concatenate([inp, pad], d=1)
       dim = inp.dim()
-    inp = dy.reshape(inp, (1, dim[0][1], dim[0][0]))
+    inp = dy.reshape(inp, (1, dim[0][1], dim[0][0]), batch_size=dim[1])
     encodings = dy.rectify(dy.conv2d_bias(inp, dy.parameter(self.filter), dy.parameter(self.bias), stride=(1, 1), is_valid=True))
     return dy.max_dim(dy.max_dim(encodings, d=1), d=0)