informative error for bad LSTM input size (#502)

* informative error for bad LSTM input size * fix unit test
neulab · Aug 3, 2018 · 2505ec5 · 2505ec5
1 parent e0f1048
commit 2505ec5
Show file tree

Hide file tree

Showing 2 changed files with 12 additions and 2 deletions.
diff --git a/test/test_run.py b/test/test_run.py
@@ -61,7 +61,7 @@ def test_segmenting(self):
   def test_reload_exception(self):
     with self.assertRaises(ValueError) as context:
       run.main(["test/config/reload_exception.yaml"])
-    self.assertEqual(str(context.exception), 'VanillaLSTMGates: x_t has inconsistent dimension')
+    self.assertEqual(str(context.exception), 'VanillaLSTMGates: x_t has inconsistent dimension 20, expecting 40')
 
   def test_report(self):
     run.main(["test/config/report.yaml"])

diff --git a/xnmt/transducers/recurrent.py b/xnmt/transducers/recurrent.py
@@ -206,9 +206,19 @@ def transduce(self, expr_seq: 'expression_seqs.ExpressionSequence') -> 'expressi
           x_t = [x_t]
         elif type(x_t) != list:
           x_t = list(x_t)
+        if sum([x_t_i.dim()[0][0] for x_t_i in x_t]) != self.input_dim:
+          found_dim = sum([x_t_i.dim()[0][0] for x_t_i in x_t])
+          raise ValueError(f"VanillaLSTMGates: x_t has inconsistent dimension {found_dim}, expecting {self.input_dim}")
         if self.dropout_rate > 0.0 and self.train:
           # apply dropout according to https://arxiv.org/abs/1512.05287 (tied weights)
-          gates_t = dy.vanilla_lstm_gates_dropout_concat(x_t, h[-1], self.Wx[layer_i], self.Wh[layer_i], self.b[layer_i], self.dropout_mask_x[layer_i], self.dropout_mask_h[layer_i], self.weightnoise_std if self.train else 0.0)
+          gates_t = dy.vanilla_lstm_gates_dropout_concat(x_t,
+                                                         h[-1],
+                                                         self.Wx[layer_i],
+                                                         self.Wh[layer_i],
+                                                         self.b[layer_i],
+                                                         self.dropout_mask_x[layer_i],
+                                                         self.dropout_mask_h[layer_i],
+                                                         self.weightnoise_std if self.train else 0.0)
         else:
           gates_t = dy.vanilla_lstm_gates_concat(x_t, h[-1], self.Wx[layer_i], self.Wh[layer_i], self.b[layer_i], self.weightnoise_std if self.train else 0.0)
         c_t = dy.vanilla_lstm_c(c[-1], gates_t)