Repository navigation
Support Float32 inputs in merge_pq - #57
Merged
Merged
Conversation
Two coupled fixes for non-Float64 factors: - The unit-2-norm check used a hardcoded 1e-12 tolerance, which is below eps(Float32). Float32-normalized columns have norm error ~1e-7 and were rejected, making merge_pq and nmfmerge unusable for Float32 data. Scale the tolerance to the input precision with sqrt(eps(eltype(W))); this only loosens the Float64 check. - The merge schedule's error field was hardcoded to Float64. Derive its element type from the factor eltypes so Float32 inputs yield Float32 merge errors matching the returned factors. 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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Two coupled fixes for non-Float64 factors:
eps(Float32). Float32-normalized columns have norm error ~1e-7 and were
rejected, making merge_pq and nmfmerge unusable for Float32 data. Scale
the tolerance to the input precision with sqrt(eps(eltype(W))); this only
loosens the Float64 check.
element type from the factor eltypes so Float32 inputs yield Float32
merge errors matching the returned factors.
Co-Authored-By: Claude Opus 4.8 noreply@anthropic.com