Give tied objects the same rank

convokit/ranker/ranker.py

@@ -48,7 +48,9 @@ def transform(
             .set_index("id")
             .sort_values(self.score_attribute_name, ascending=False)
         )
-        df[self.rank_attribute_name] = [idx + 1 for idx, _ in enumerate(df.index)]
+        df[self.rank_attribute_name] = (
+            df[self.score_attribute_name].rank(ascending=False, method="min").astype(int)
+        )
 
         for obj in corpus.iter_objs(obj_type=self.obj_type):
             if obj.id in df.index:
@@ -72,7 +74,9 @@ def transform_objs(self, objs: List[CorpusComponent]):
             .set_index("id")
             .sort_values(self.score_attribute_name, ascending=False)
         )
-        df[self.rank_attribute_name] = [idx + 1 for idx, _ in enumerate(df.index)]
+        df[self.rank_attribute_name] = (
+            df[self.score_attribute_name].rank(ascending=False, method="min").astype(int)
+        )
         for obj in objs:
             obj.add_meta(self.score_attribute_name, df.loc[obj.id][self.score_attribute_name])
             obj.add_meta(self.rank_attribute_name, df.loc[obj.id][self.rank_attribute_name])

convokit/tests/ranker/test_ranker.py

@@ -0,0 +1,62 @@
+import unittest
+
+from convokit import Corpus, Speaker, Utterance
+from convokit.ranker import Ranker
+
+
+def construct_tied_corpus():
+    """Four utterances whose scores contain a three-way and a two-way tie."""
+    scores = {"0": 10, "1": 10, "2": 5, "3": 10}
+    return Corpus(
+        utterances=[
+            Utterance(
+                id=uid,
+                text=f"utterance {uid}",
+                speaker=Speaker(id=f"s{uid}"),
+                meta={"score": score},
+            )
+            for uid, score in scores.items()
+        ]
+    )
+
+
+class TestRankerTies(unittest.TestCase):
+    def setUp(self):
+        self.corpus = construct_tied_corpus()
+        self.ranker = Ranker(
+            obj_type="utterance",
+            score_func=lambda utt: utt.meta["score"],
+            score_attribute_name="score_value",
+            rank_attribute_name="rank_value",
+        )
+
+    def test_transform_gives_tied_objects_the_same_rank(self):
+        self.ranker.transform(self.corpus)
+        ranks = {utt.id: utt.meta["rank_value"] for utt in self.corpus.iter_utterances()}
+        # Three utterances score 10 and one scores 5. Competition ranking gives the
+        # tied objects the same rank and skips the ones they consumed.
+        assert ranks["0"] == ranks["1"] == ranks["3"] == 1
+        assert ranks["2"] == 4
+
+    def test_transform_objs_gives_tied_objects_the_same_rank(self):
+        objs = list(self.corpus.iter_utterances())
+        self.ranker.transform_objs(objs)
+        ranks = {utt.id: utt.meta["rank_value"] for utt in objs}
+        assert ranks["0"] == ranks["1"] == ranks["3"] == 1
+        assert ranks["2"] == 4
+
+    def test_distinct_scores_still_rank_consecutively(self):
+        corpus = Corpus(
+            utterances=[
+                Utterance(id="a", text="a", speaker=Speaker(id="sa"), meta={"score": 3}),
+                Utterance(id="b", text="b", speaker=Speaker(id="sb"), meta={"score": 2}),
+                Utterance(id="c", text="c", speaker=Speaker(id="sc"), meta={"score": 1}),
+            ]
+        )
+        self.ranker.transform(corpus)
+        ranks = {utt.id: utt.meta["rank_value"] for utt in corpus.iter_utterances()}
+        assert ranks == {"a": 1, "b": 2, "c": 3}
+
+
+if __name__ == "__main__":
+    unittest.main()