Added assume_sorted argument

rtrec/recommender.py

@@ -35,6 +35,7 @@ def fit(
         batch_size: int = 1_000,
         update_interaction: bool = False,
         parallel: bool = False,
+        assume_sorted: bool = True
     ) -> Self:
         """
         Fit the recommender model on the given DataFrame of interactions.
@@ -44,6 +45,7 @@ def fit(
         :param batch_size (int): The number of interactions per mini-batch. Defaults to 1000.
         :param update_interaction (bool): Whether to update existing interactions. Defaults to False.
         :param parallel (bool): Whether to run the fitting process in parallel. Defaults to False.
+        :param assume_sorted (bool): Whether the interactions are already sorted by timestamp. Defaults to True.
         """
         start_time = time.time()
 
@@ -57,6 +59,9 @@ def fit(
 
         # Add interactions to the model
         interaction_df = train_data[["user", "item", "tstamp", "rating"]]
+        if not assume_sorted:
+            # sort interactions by timestamp ascending order
+            interaction_df.sort_values("tstamp", ascending=True, inplace=True)
         total = math.ceil(len(interaction_df) / batch_size)
         for batch in tqdm(generate_batches(interaction_df, batch_size, as_generator=self.use_generator), total=total, desc="Add interactions"):
             self.model.add_interactions(batch, update_interaction=update_interaction, record_interactions=True)
@@ -74,7 +79,8 @@ def bulk_fit(
             item_tags: Optional[Dict[Any, List[str]]] = None,
             batch_size: int = 1_000,
             update_interaction: bool=False,
-            parallel: bool=True
+            parallel: bool=True,
+            assume_sorted: bool=True
     ) -> Self:
         """
         Fit the recommender model on the given DataFrame of interactions in a single batch.
@@ -84,6 +90,7 @@ def bulk_fit(
         :param batch_size (int): The number of interactions per mini-batch. Defaults to 1000.
         :param update_interaction (bool): Whether to update existing interactions. Defaults to False.
         :param parallel (bool): Whether to run the fitting process in parallel. Defaults to True.
+        :param assume_sorted (bool): Whether the interactions are already sorted by timestamp. Defaults to True.
         """
         start_time = time.time()
 
@@ -97,6 +104,9 @@ def bulk_fit(
 
         # Add interactions to the model
         interaction_df = train_data[["user", "item", "tstamp", "rating"]]
+        if not assume_sorted:
+            # sort interactions by timestamp ascending order inplace
+            interaction_df.sort_values("tstamp", ascending=True, inplace=True)
         total = math.ceil(len(interaction_df) / batch_size)
         for batch in tqdm(generate_batches(interaction_df, batch_size, as_generator=self.use_generator), total=total, desc="Add interactions"):
             self.model.add_interactions(batch, update_interaction=update_interaction)