fix bug in SeqLabelTester

fastNLP/core/tester.py

@@ -59,6 +59,8 @@ class BaseTester(object):
                 self.batch_output.append(prediction)
             if self.save_loss:
                 self.eval_history.append(eval_results)
+            if step % n_print == 0:
+                print('[test step: {:>4}]'.format(step))
             step += 1
 
     def prepare_input(self, data_path):
@@ -134,7 +136,7 @@ class SeqLabelTester(BaseTester):
         results = torch.Tensor(prediction).view(-1,)
         # make sure "results" is in the same device as "truth"
         results = results.to(truth)
-        accuracy = torch.sum(results == truth.view((-1,))) / results.shape[0]
+        accuracy = torch.sum(results == truth.view((-1,))).to(torch.float) / results.shape[0]
         return [loss.data, accuracy.data]
 
     def metrics(self):
@@ -153,7 +155,6 @@ class SeqLabelTester(BaseTester):
     def make_batch(self, iterator, data):
         return Action.make_batch(iterator, data, use_cuda=self.use_cuda, output_length=True)
 
-
 class ClassificationTester(BaseTester):
     """Tester for classification."""