Revised to filter row at to_csr()

rtrec/models/internal/slim_elastic.py

@@ -358,9 +358,8 @@ def recommend(self, user_id: int, interaction_matrix: sp.csr_matrix, candidate_i
         # Get the top-K items by sorting the predicted scores in descending order
         # [::-1] reverses the order to get the items with the highest scores first
         top_items = np.argsort(user_scores)[-top_k:][::-1]
-
         return top_items
-    
+
     def similar_items(self, item_id: int, top_k: int=10):
         """
         Get the top-K most similar items to a given item.

rtrec/models/slim.py

@@ -43,7 +43,7 @@ def _recommend(self, user_id: int, candidate_item_ids: List[int], top_k: int = 1
         :param filter_interacted: Whether to filter out items the user has already interacted with
         :return: List of top-K item indices recommended for the user
         """
-        interaction_matrix = self.interactions.to_csr()
+        interaction_matrix = self.interactions.to_csr(select_users=[user_id])
         return self.model.recommend(user_id, interaction_matrix, candidate_item_ids, top_k=top_k, filter_interacted=filter_interacted)
 
     def _similar_items(self, query_item_id: int, top_k: int = 10) -> List[int]:

rtrec/utils/interactions.py

@@ -171,17 +171,19 @@ def get_all_non_negative_items(self, user_id: int) -> List[int]:
         return [item_id for item_id in self.all_item_ids
                 if self.get_user_item_rating(user_id, item_id, default_rating=0.0) >= 0.0]
 
-    def to_csr(self) -> csr_matrix:
+    def to_csr(self, select_users: List[int] = None) -> csr_matrix:
         rows, cols, data = [], [], []
         max_row, max_col = 0, 0
 
         for user, inner_dict in self.interactions.items():
             for item, (rating, tstamp) in inner_dict.items():
+                max_row = max(max_row, user)
+                max_col = max(max_col, item)
+                if select_users is not None and user not in select_users:
+                    continue
                 rows.append(user)
                 cols.append(item)
                 data.append(self._apply_decay(rating, tstamp))
-                max_row = max(max_row, user)
-                max_col = max(max_col, item)
 
         # Create the csr_matrix
         return csr_matrix((data, (rows, cols)), shape=(max_row + 1, max_col + 1))