Skip to content

Fix helm_normalizer mis-scoring numbers (homogeneize before remove_punc) #1273

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
iamsharduld wants to merge 1 commit into
base: main
Choose a base branch
from
Open

Fix helm_normalizer mis-scoring numbers (homogeneize before remove_punc) #1273

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
4 changes: 3 additions & 1 deletion src/lighteval/metrics/normalizations.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -68,7 +68,9 @@ def lower(text: str) -> str:
def _tokenize(text):
return re.split(" |-", text)

tokens = [white_space_fix(remove_articles(homogeneize_numbers(remove_punc(lower(t))))) for t in _tokenize(text)]
# homogeneize_numbers must run before remove_punc: otherwise remove_punc strips the decimal
# point first, so "1.0" -> "10" -> float -> "10.0" (distinct numbers collide, and 1.0 != 1).
tokens = [white_space_fix(remove_articles(remove_punc(homogeneize_numbers(lower(t))))) for t in _tokenize(text)]
return " ".join([t for t in tokens if t != ""]).strip()


Expand Down
9 changes: 9 additions & 0 deletions tests/test_unit_base_metrics.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -95,6 +95,15 @@ def test_quasi_exact_match(self):
res = em.compute_one_item("", "")
assert res == 0

def test_quasi_exact_match_numbers(self):
em = ExactMatches(normalize_gold=helm_normalizer, normalize_pred=helm_normalizer)

# Numbers that are equal but formatted differently must match (homogeneize_numbers' goal).
assert em.compute_one_item("1.0", "1") == 1
# Distinct numbers must NOT be scored as an exact match.
assert em.compute_one_item("10", "1.0") == 0
assert em.compute_one_item("3.14", "314") == 0

def test_prefix_exact_match(self):
em = ExactMatches(
strip_strings=True,
Expand Down