blakeslee-research/nucleic-purpose-classifier
3
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Merge nucleic/plucky-north-vole-sdna into dev

Browse Source
This commit is contained in:
abkslm
2026-08-01 06:29:50 -07:00
parent 2ed9d28c28
commit bb4dff9b56
5 changed files with 161 additions and 112 deletions
Download Patch File Download Diff File Expand all files Collapse all files
tests/test_export_swe_chat.py
+20 -20
View File
@@ -26,16 +26,15 @@ def row(session, turn, prompt, **overrides):
class ExportSWEChatTests(unittest.TestCase):
def test_selects_first_three_orders_dedupes_and_caps_sources(self):
def test_selects_first_prompt_and_its_response_dedupes_and_caps_sources(self):
rows = [
row("s1", "t3", "third", conversation_turn_number=3),
row("s1", "t1", "first", conversation_turn_number=1),
row("s1", "t2", "second", conversation_turn_number=2),
row("s2", "t1", " first ", conversation_turn_number=1),
row("s2", "t2", "later", conversation_turn_number=2),
row("s2", "t3", "later again", conversation_turn_number=3),
row("s3", "t1", "one"), row("s3", "t2", "two"),
row("s4", "t1", "another one"), row("s4", "t2", "another two"), row("s4", "t3", "another three"),
row("s1", "t2", "answer", conversation_turn_number=1, turn_type="assistant_response", role="assistant"),
row("s1", "t1", "first", conversation_turn_number=0),
row("s2", "t1", " first ", conversation_turn_number=0),
row("s2", "t2", "answer", conversation_turn_number=1, turn_type="assistant_response", role="assistant"),
row("s3", "t1", "one"),
row("s4", "t1", "another one", conversation_turn_number=0),
row("s4", "t2", "answer", conversation_turn_number=1, turn_type="assistant_response", role="assistant"),
]
candidates, funnel = export_swe_chat.select_candidates(
rows,
@@ -43,30 +42,31 @@ class ExportSWEChatTests(unittest.TestCase):
max_per_repo=1, max_per_user=1,
)
self.assertEqual(1, len(candidates))
self.assertEqual(["t1", "t2", "t3"], [turn.turn_id for turn in candidates[0].turns])
self.assertEqual("first", candidates[0].turns[0].prompt)
self.assertEqual(["t1", "t2"], [candidates[0].prompt_turn.turn_id, candidates[0].response_turn.turn_id])
self.assertEqual("first", candidates[0].prompt_turn.prompt)
self.assertEqual(1, funnel["rejectedDuplicateFirstPrompt"])
self.assertEqual(1, funnel["sessionsFewerThanThreeEligiblePrompts"])
self.assertEqual(1, funnel["sessionsWithoutFirstAssistantResponse"])
self.assertEqual(1, funnel["rejectedRepoCap"])
def test_filters_non_user_continuation_empty_and_ambiguous_ordinals(self):
def test_filters_non_user_continuation_empty_and_missing_response(self):
rows = [
row("s1", "t1", "x", role="assistant"), row("s1", "t2", "x", is_continuation=True), row("s1", "t3", ""),
row("s2", "t1", "a", conversation_turn_number=1, turn_number=1), row("s2", "t2", "b", conversation_turn_number=1, turn_number=1), row("s2", "t3", "c", conversation_turn_number=3),
row("s2", "t1", "a", conversation_turn_number=0), row("s2", "t2", "wrong ordinal", conversation_turn_number=2, turn_type="assistant_response", role="assistant"),
]
candidates, funnel = export_swe_chat.select_candidates(rows, sessions={}, max_per_repo=10, max_per_user=10)
self.assertEqual([], candidates)
self.assertEqual(1, funnel["rejectedRole"])
self.assertEqual(1, funnel["rejectedContinuation"])
self.assertEqual(1, funnel["rejectedMalformedOrEmpty"])
self.assertEqual(1, funnel["sessionsAmbiguousTurnOrder"])
self.assertEqual(1, funnel["rejectedUserMalformedOrEmpty"])
self.assertEqual(1, funnel["sessionsWithoutFirstAssistantResponse"])
def test_candidate_only_retains_first_prompt_as_student_text(self):
turns = tuple(export_swe_chat.Turn("s", f"t{index}", index, index, prompt) for index, prompt in enumerate(("first", "second", "third"), start=1))
value = export_swe_chat.Candidate("s", "r", "u", turns).json("a" * 40)
prompt = export_swe_chat.Turn("s", "t1", 0, 0, "first")
response = export_swe_chat.Turn("s", "t2", 1, 1, "answer")
value = export_swe_chat.Candidate("s", "r", "u", prompt, response).json("a" * 40)
self.assertEqual("first", value["prompt"])
self.assertEqual(["second", "third"], value["teacherContext"])
self.assertNotIn("second", value["contextPromptHashes"])
self.assertEqual("answer", value["teacherResponse"])
self.assertNotIn("answer", value["teacherResponseHash"])
if __name__ == "__main__":