Bug fix: use max_length instead of true length for seq lengths

boraerden · Apr 18, 2018 · 374b725 · 374b725
1 parent 25e3c34
commit 374b725
Showing 1 changed file with 2 additions and 1 deletion.
diff --git a/tf_rnn_classifier.py b/tf_rnn_classifier.py
@@ -251,7 +251,8 @@ def _convert_X(self, X):
         index = dict(zip(self.vocab, range(len(self.vocab))))
         unk_index = index['$UNK']
         for i in range(new_X.shape[0]):
-            ex_lengths.append(len(X[i]))
+            ex_len = min([len(X[i]), self.max_length])
+            ex_lengths.append(ex_len)
             vals = X[i][-self.max_length: ]
             vals = [index.get(w, unk_index) for w in vals]
             temp = np.zeros((self.max_length,), dtype='int')