Merge pull request #236 from xionghhcs/master

fix the tl.utils.predict 's bug. when the data size can not be exactly divided by batch_size, there will be some predict result lost.  by @xionghhcs

Author: zsdonghao

tensorlayer/utils.py

@@ -281,7 +281,21 @@ def predict(sess, network, X, x, y_op, batch_size=None):
             if result is None:
                 result = result_a
             else:
-                result = np.hstack((result, result_a))
+                result = np.vstack((result, result_a))
+        if result is None:
+            if len(X) % batch_size != 0:
+                dp_dict = dict_to_one(network.all_drop)
+                feed_dict = {x:  X[-(len(X) % batch_size):, :], }
+                feed_dict.update(dp_dict)
+                result_a = sess.run(y_op, feed_dict=feed_dict)
+                result = result_a
+        else:
+            if len(X) != len(result) and len(X) % batch_size != 0:
+                dp_dict = dict_to_one(network.all_drop)
+                feed_dict = {x: X[-(len(X) % batch_size):, :], }
+                feed_dict.update(dp_dict)
+                result_a = sess.run(y_op, feed_dict=feed_dict)
+                result = np.vstack((result, result_a))
         return result