bugfix: Ranking value from suggestions are not properly converted from HF datasets

CHANGELOG.md

@@ -19,6 +19,7 @@ These are the section headers that we use:
 ### Fixed
 
 - Fixed prepare for training when passing `RankingValueSchema` instances to suggestions. ([#4628](https://github.com/argilla-io/argilla/pull/4628))
+- Fixed parsing ranking values in suggestions from HF datasets. ([#4629](https://github.com/argilla-io/argilla/pull/4629))
 
 ## [1.25.0](https://github.com/argilla-io/argilla/compare/v1.24.0...v1.25.0)
 

src/argilla/client/feedback/integrations/huggingface/dataset.py

@@ -428,10 +428,11 @@ def from_huggingface(
                     f"{question.name}-suggestion" in hfds[index]
                     and hfds[index][f"{question.name}-suggestion"] is not None
                 ):
-                    suggestion = {
-                        "question_name": question.name,
-                        "value": hfds[index][f"{question.name}-suggestion"],
-                    }
+                    value = hfds[index][f"{question.name}-suggestion"]
+                    if question.type == QuestionTypes.ranking:
+                        value = [{"rank": r, "value": v} for r, v in zip(value["rank"], value["value"])]
+
+                    suggestion = {"question_name": question.name, "value": value}
                     if hfds[index][f"{question.name}-suggestion-metadata"] is not None:
                         suggestion.update(hfds[index][f"{question.name}-suggestion-metadata"])
                     suggestions.append(suggestion)

src/argilla/client/feedback/schemas/questions.py

@@ -274,7 +274,7 @@ class RankingQuestion(QuestionSchema, LabelMappingMixin):
 
     Examples:
         >>> from argilla.client.feedback.schemas.questions import RankingQuestion
-        >>> RankingQuestion(name="ranking_question", title="Ranking Question", labels=["label_1", "label_2"])
+        >>> RankingQuestion(name="ranking_question", title="Ranking Question", values=["label_1", "label_2"])
     """
 
     type: Literal[QuestionTypes.ranking] = Field(QuestionTypes.ranking.value, allow_mutation=False)

tests/integration/client/feedback/dataset/local/test_dataset.py

@@ -566,6 +566,10 @@ def test_push_to_huggingface_and_from_huggingface(
             hf_response.dict() == response.dict()
             for hf_response, response in zip(hf_record.responses, record.responses)
         )
+        assert all(
+            hf_suggestion.dict() == suggestion.dict()
+            for hf_suggestion, suggestion in zip(hf_record.suggestions, record.suggestions)
+        ), f"{[s.dict() for s in hf_record.suggestions]} != {[s.dict() for s in record.suggestions]}"
 
     dataset.add_records(
         records=[