Feature/safe random negatives (#161)

* bugfix

* updated RAGTrainer data handling

* remove unnec changes

* add back 02 example notebook

* only select as many negatives as you can

* remove old changes

* remove unnecessary changes

---------

Co-authored-by: Benjamin Clavié <ben@clavie.eu>

ragatouille/data/training_data_processor.py

@@ -125,7 +125,7 @@ def _get_new_negatives(self, query, passages, mine_hard_negatives, n_new_negativ
         else:
             new_negatives = [
                 x
-                for x in random.sample(self.collection, n_new_negatives)
+                for x in random.sample(self.collection, min(n_new_negatives, len(self.collection)))
                 if x not in passages["positives"] and x not in passages["negatives"]
             ]