Skip to content

Fix graph Laplacian docstring and semantic weight scaling in KernelLanguageEntropy #441

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
Kangwei-g wants to merge 1 commit into
base: main
Choose a base branch
from
Open

Fix graph Laplacian docstring and semantic weight scaling in KernelLanguageEntropy #441

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
8 changes: 4 additions & 4 deletions src/lm_polygraph/estimators/kernel_language_entropy.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -94,7 +94,7 @@ def __call__(self, stats: Dict[str, np.ndarray]) -> np.ndarray:
2. Let NLI'(Si, Sj) = one-hot prediction over (entailment, neutral class, contradiction)
Note that NLI'(Si, Sj) is calculated in stats
3. Let W be a matrix, such that Wij = wNLI'(Si, Sj), where w = (1, 0.5, 0)
4. Let L be a laplacian matrix of W, i.e. L = W - D, where Dii = sum(Wij) over j.
4. Let L be a laplacian matrix of W, i.e. L = D - W, where Dii = sum(Wij) over j.
5. Let Kheat = heat kernel of W, i.e. Kheat = expm(-t * L), where t is a hyperparameter.
6. Finally, KLE(x) = VNE(Kheat), where VNE(A) = -Tr(A log A).
"""
Expand All @@ -105,11 +105,11 @@ def __call__(self, stats: Dict[str, np.ndarray]) -> np.ndarray:
for matrix_entail, matrix_contra in zip(
semantic_matrix_entail, semantic_matrix_contra
):
matrix_entail = (matrix_entail + matrix_entail.T) / 2
matrix_contra = (matrix_contra + matrix_contra.T) / 2
matrix_entail = (matrix_entail + matrix_entail.T)
matrix_contra = (matrix_contra + matrix_contra.T)

matrix_neutral = (
np.ones(matrix_entail.shape) - matrix_entail - matrix_contra
2 * np.ones(matrix_entail.shape) - matrix_entail - matrix_contra
)
weighted_graph = matrix_entail + 0.5 * matrix_neutral

Expand Down
Loading