Merge nucleic/sleek-ember-seal-uady into dev
This commit is contained in:
@@ -23,5 +23,33 @@ class TierDriftTests(unittest.TestCase):
|
||||
self.assertLessEqual(report["maximumTierDrift"], 1)
|
||||
|
||||
|
||||
class PredictionAgreementTests(unittest.TestCase):
|
||||
def test_reports_accuracy_transitions_and_scorable_agreement(self):
|
||||
records = [
|
||||
{"prompt": "one", "slice": "core"},
|
||||
{"prompt": "two", "slice": "boundary"},
|
||||
{"prompt": "three", "slice": "vague-eval"},
|
||||
{"prompt": "four", "slice": "core"},
|
||||
]
|
||||
report = purpose_eval.prediction_agreement(
|
||||
records,
|
||||
actual=[0, 1, 2, 3],
|
||||
reference=[0, 0, 3, 4],
|
||||
candidate=[1, 1, 4, 5],
|
||||
)
|
||||
self.assertEqual(0.0, report["labelAgreement"])
|
||||
self.assertEqual(0.0, report["scoredLabelAgreement"])
|
||||
self.assertEqual(
|
||||
{
|
||||
"correctToIncorrect": 1,
|
||||
"differentIncorrectLabel": 2,
|
||||
"incorrectToCorrect": 1,
|
||||
},
|
||||
report["transitionCounts"],
|
||||
)
|
||||
self.assertEqual(2, report["bySlice"]["core"]["records"])
|
||||
self.assertEqual(4, len(report["disagreements"]))
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
unittest.main()
|
||||
|
||||
Reference in New Issue
Block a user