Repository navigation
Add a certified multiple-minima example - #63
Merged
Merged
Conversation
Introduce a 4x8 rank-3 integer matrix on which rank-3 NMF has a spurious local minimum: standard NMF from NNDSVD converges to relative error ~7e-4, while over-factoring to 4 components and merging back to 3 recovers the exact factorization. The test certifies the spurious point is a strict local minimum (modulo column scaling) rather than slow convergence: KKT holds with positive dual variables and the reduced Hessian is positive definite off the scaling nullspace. `certify_localmin` finite-differences the analytic gradient on the free coordinates. Use the example as the documentation's headline motivating example. Remove the previous example from both the README and the docs. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Introduce a 4x8 rank-3 integer matrix on which rank-3 NMF has a spurious local minimum: standard NMF from NNDSVD converges to relative error ~7e-4, while over-factoring to 4 components and merging back to 3 recovers the exact factorization.
The test certifies the spurious point is a strict local minimum (modulo column scaling) rather than slow convergence: KKT holds with positive dual variables and the reduced Hessian is positive definite off the scaling nullspace.
certify_localminfinite-differences the analytic gradient on the free coordinates.Use the example as the documentation's headline motivating example. Remove the previous example from both the README and the docs.