Skip to content

[BUG] Iterative LLPR regularization is not working as intended #1154

Description

@bananenpampe

Summary

It appears like the LLPR regularization is not working as intended.
The intended behaviour should be that a small regularization term that makes the matrix invertible is added to the covariance matrix of the last layer features F^T.F.
I find having to manually set the regularization strengths via a grid search, otherwise I obain an invertability error.

_LinAlgError: linalg.cholesky: The factorization could not be completed because the input is not positive-definite (the leading minor of order 43 is not positive-definite).

Expected behavior

LLPR wrapper exports an LLPR enabled model.

Actual behavior

_LinAlgError: linalg.cholesky: The factorization could not be completed because the input is not positive-definite (the leading minor of order 43 is not positive-definite).

Version

2026.2.1

Steps to reproduce

No response

Further information, files, and links

No response

Metadata

Metadata

Assignees

No one assigned

    Labels

    BugSomething isn't working

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions