Skip to content

TensorKitchen Patch - #24

Merged
sChoiKr merged 19 commits into
mainfrom
default-convergence-gradient-tolerance
Jun 18, 2026
Merged

TensorKitchen Patch#24
sChoiKr merged 19 commits into
mainfrom
default-convergence-gradient-tolerance

Conversation

@sChoiKr

@sChoiKr sChoiKr commented Jun 17, 2026

Copy link
Copy Markdown
Contributor

No description provided.

@PBrdng PBrdng changed the title Default convergence gradient tolerance TensorKitchen Patch Jun 17, 2026
@PBrdng

PBrdng commented Jun 17, 2026

Copy link
Copy Markdown
Member

This patch adds a few improvements:

  • LFBGS is now working correctly
  • We now optimize the relative error $0.5 \cdot \Vert A - X \Vert^2 / \Vert A\Vert^2$ by default, where $X$ is the variable we optimize.
  • Repeated code was unified and cleaned up.
  • Helper functions were moved to a separate place in the code
  • Progress Meter now prints the second phase always if the first phase was printed.

@PBrdng

PBrdng commented Jun 17, 2026

Copy link
Copy Markdown
Member

Nice: this works well now

a, b, c, d = 100, 50, 20, 200
A = reshape(sum(kron(randn(d), randn(c), randn(b), randn(a)) for _ in 1:5), a, b, c, d);
cpd(A, 5)

@TensorKitchen TensorKitchen deleted a comment from sChoiKr Jun 17, 2026
@sChoiKr
sChoiKr merged commit 7973303 into main Jun 18, 2026
3 checks passed
@sChoiKr
sChoiKr deleted the default-convergence-gradient-tolerance branch June 18, 2026 00:38
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants