Modified predict to compute representations just once

Author: paoloRais

lightfm/lightfm.py

@@ -562,7 +562,7 @@ def predict(self, user_ids, item_ids, item_features=None,
              Number of parallel computation threads to use. Should
              not be higher than the number of physical cores.
         precompute_representations: boolean
-            if True, the representations are pre-computed for all items 
+            if True, the representations are pre-computed for all items
             and all users, instead of being computed once for each pair.
 
         Returns