Adjust hessian_reg to fix Arnoldi test case

pytorch / captum

Model interpretability and understanding for PyTorch

https://captum.ai

BSD 3-Clause "New" or "Revised" License

4.96k stars 499 forks source link

Adjust hessian_reg to fix Arnoldi test case #1388

Closed vivekmig closed 1 month ago

vivekmig commented 1 month ago

Summary: Increase hessian regularization term for Arnoldi test to ensure that highest self-influence datapoint / proponent match the corresponding example

Differential Revision: D64584502

facebook-github-bot commented 1 month ago

This pull request was exported from Phabricator. Differential Revision: D64584502

facebook-github-bot commented 1 month ago

This pull request has been merged in pytorch/captum@75be44f9fbf58ec2c104aa2d911fa633b0fa1f7a.