Allow excluding train set positives from test set evaluation.

lightfm/_lightfm_fast.pyx.template

@@ -1188,7 +1188,8 @@ def predict_lightfm(CSRMatrix item_features,
 
 def predict_ranks(CSRMatrix item_features,
                   CSRMatrix user_features,
-                  CSRMatrix interactions,
+                  CSRMatrix test_interactions,
+                  CSRMatrix train_interactions,
                   flt[::1] ranks,
                   FastLightFM lightfm,
                   int num_threads):
@@ -1205,10 +1206,10 @@ def predict_ranks(CSRMatrix item_features,
 
     # Figure out the max size of the predictions
     # buffer.
-    for user_id in range(interactions.rows):
+    for user_id in range(test_interactions.rows):
         predictions_size = int_max(predictions_size,
-                                   interactions.get_row_end(user_id)
-                                   - interactions.get_row_start(user_id))
+                                   test_interactions.get_row_end(user_id)
+                                   - test_interactions.get_row_start(user_id))
 
     {nogil_block}
 
@@ -1217,7 +1218,7 @@ def predict_ranks(CSRMatrix item_features,
         item_ids = <int *>malloc(sizeof(int) * predictions_size)
         predictions = <flt *>malloc(sizeof(flt) * predictions_size)
 
-        for user_id in {range_block}(interactions.rows):
+        for user_id in {range_block}(test_interactions.rows):
 
             compute_representation(user_features,
                                    lightfm.user_features,
@@ -1227,14 +1228,14 @@ def predict_ranks(CSRMatrix item_features,
                                    lightfm.user_scale,
                                    user_repr)
 
-            row_start = interactions.get_row_start(user_id)
-            row_stop = interactions.get_row_end(user_id)
+            row_start = test_interactions.get_row_start(user_id)
+            row_stop = test_interactions.get_row_end(user_id)
 
             # Compute predictions for the items whose
             # ranks we want to know
             for i in range(row_stop - row_start):
 
-                item_id = interactions.indices[row_start + i]
+                item_id = test_interactions.indices[row_start + i]
 
                 compute_representation(item_features,
                                        lightfm.item_features,
@@ -1250,7 +1251,10 @@ def predict_ranks(CSRMatrix item_features,
                                                               lightfm.no_components)
 
             # Now we can zip through all the other items and compute ranks
-            for item_id in range(interactions.cols):
+            for item_id in range(test_interactions.cols):
+
+                if in_positives(item_id, user_id, train_interactions):
+                    continue
 
                 compute_representation(item_features,
                                        lightfm.item_features,
@@ -1273,6 +1277,7 @@ def predict_ranks(CSRMatrix item_features,
 
 
 def calculate_auc_from_rank(CSRMatrix ranks,
+                            int[::1] num_train_positives,
                             flt[::1] rank_data,
                             flt[::1] auc,
                             int num_threads):
@@ -1287,7 +1292,7 @@ def calculate_auc_from_rank(CSRMatrix ranks,
             row_stop = ranks.get_row_end(user_id)
 
             num_positives = row_stop - row_start
-            num_negatives = ranks.cols - (row_stop - row_start)
+            num_negatives = ranks.cols - ((row_stop - row_start) + num_train_positives[user_id])
 
             # If there is only one class present,
             # return 0.5.