Skip to content

Fix Probability metric to aggregate over gold choices, not all choices #1461

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
Sivadheeraj5356 wants to merge 2 commits into
base: main
Choose a base branch
from
Open

Fix Probability metric to aggregate over gold choices, not all choices #1461

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
0edce82
Fix Probability metric to aggregate gold choice probabilities
Sivadheeraj5356 Oct 31, 2026
c4df9d7
Update Probability metric tests for gold choice aggregation
Sivadheeraj5356 Oct 31, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
3 changes: 2 additions & 1 deletion src/lighteval/metrics/metrics_sample.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -402,7 +402,8 @@ def compute(
else logprobs
)
probs = np.exp(normalized_log_probs)
return self.aggregation_function(probs)
gold_ixs = as_list(doc.gold_index)
return self.aggregation_function(probs[gold_ixs])


class Recall(SampleLevelComputation):
Expand Down
13 changes: 10 additions & 3 deletions tests/test_unit_base_metrics.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -200,10 +200,17 @@ def test_prob(self):
result = prob_metric.compute_sample(doc=doc, model_response=model_response)
assert result[prob_metric.metric_name] == pytest.approx(0.7)

# Aggregation function test
model_response = ModelResponse(logprobs=np.log([0.7, 0.1]))
# Aggregation function test over multiple golds
doc_multi_gold = Doc(query="Test query", choices=["A", "B", "C"], gold_index=[0, 1], task_name="test")
model_response = ModelResponse(logprobs=np.log([0.7, 0.1, 0.05]))
prob_min_metric = ProbabilityMetric(aggregation_function=np.min)
result = prob_min_metric.compute_sample(doc=doc, model_response=model_response)
result = prob_min_metric.compute_sample(doc=doc_multi_gold, model_response=model_response)
assert result[prob_metric.metric_name] == pytest.approx(0.1)

# Gold is not the argmax: the metric must return the gold probability
doc_gold_not_argmax = Doc(query="Test query", choices=["A", "B", "C"], gold_index=1, task_name="test")
model_response = ModelResponse(logprobs=np.log([0.7, 0.1, 0.05]))
result = prob_metric.compute_sample(doc=doc_gold_not_argmax, model_response=model_response)
assert result[prob_metric.metric_name] == pytest.approx(0.1)

def test_mc_probability_metric(self):
Expand Down