diff --git a/ClickThroughRate/WideDeepLearning/wdl_train_eval.py b/ClickThroughRate/WideDeepLearning/wdl_train_eval.py
index be3fe7c785dc6ebdc5a56a208a76fca5bbbac570..ee243cce60158defa705bb669eea4be8611cc39a 100644
--- a/ClickThroughRate/WideDeepLearning/wdl_train_eval.py
+++ b/ClickThroughRate/WideDeepLearning/wdl_train_eval.py
@@ -182,9 +182,9 @@ def main():
             eval_loss = 0.0
             for j in range(FLAGS.eval_batchs):
                 loss, pred, ref = eval_job().get()
-                label_ = ref.ndarray().astype(np.float32)
+                label_ = ref.numpy().astype(np.float32)
                 labels = np.concatenate((labels, label_), axis=0)
-                preds = np.concatenate((preds, pred.ndarray()), axis=0)
+                preds = np.concatenate((preds, pred.numpy()), axis=0)
                 eval_loss += loss.mean()
             auc = roc_auc_score(labels[1:], preds[1:])
             print(i+1, "eval_loss", eval_loss/FLAGS.eval_batchs, "eval_auc", auc)
diff --git a/ClickThroughRate/WideDeepLearning/wdl_train_eval_test.py b/ClickThroughRate/WideDeepLearning/wdl_train_eval_test.py
index f9d52644491099c0bfbe417465ca321f722afafe..862d64944ab316dce6eddbf7744363fca4707cd8 100644
--- a/ClickThroughRate/WideDeepLearning/wdl_train_eval_test.py
+++ b/ClickThroughRate/WideDeepLearning/wdl_train_eval_test.py
@@ -205,9 +205,9 @@ def main():
         eval_loss = 0.0
         for i in range(eval_epoch_size):
             loss, pred, ref = eval_job().get()
-            label_ = ref.ndarray().astype(np.float32)
+            label_ = ref.numpy().astype(np.float32)
             labels = np.concatenate((labels, label_), axis=0)
-            preds = np.concatenate((preds, pred.ndarray()), axis=0)
+            preds = np.concatenate((preds, pred.numpy()), axis=0)
             eval_loss += loss.mean()
         auc = roc_auc_score(labels[1:], preds[1:])
         print(epoch, "eval_loss", eval_loss/eval_epoch_size, "eval_auc", auc)
@@ -217,9 +217,9 @@ def main():
     eval_loss = 0.0
     for i in range(test_epoch_size):
         loss, pred, ref = test_job().get()
-        label_ = ref.ndarray().astype(np.float32)
+        label_ = ref.numpy().astype(np.float32)
         labels = np.concatenate((labels, label_), axis=0)
-        preds = np.concatenate((preds, pred.ndarray()), axis=0)
+        preds = np.concatenate((preds, pred.numpy()), axis=0)
         eval_loss += loss.mean()
     auc = roc_auc_score(labels[1:], preds[1:])
     print("test_loss", eval_loss/test_epoch_size, "eval_auc", auc)