-
Notifications
You must be signed in to change notification settings - Fork 52
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Stable hyperbolic functions #631
Conversation
Codecov Report
@@ Coverage Diff @@
## master #631 +/- ##
==========================================
- Coverage 99.01% 99.01% -0.01%
==========================================
Files 104 104
Lines 10138 10137 -1
==========================================
- Hits 10038 10037 -1
Misses 100 100
📣 We’re building smart automated test selection to slash your CI/CD build times. Learn more |
Improving accuracy for |
We also noticed that parallel transport might be more stable in a different implementation. Actually the ideas come from Manopt/Matlab, where they are documented as more stable – but then where would we put such an accuracy comparison? A small quarto where we define and compare both variants? |
Thanks for your remarks Mateusz :) I hacked a small start of a comparison; I can tell from that, that both implementations yield different results, but since I use the old one for the BigFloat, the old one has a smaller error than the new one. So I am not so sure what to compare to. edit: But if I would see how to compare these, for one function, I can happily do the other functions as well. Especially also to look where the Manopt/Matlab code states that the current implementation we have might be more stable for larger tangent vectors (in norm), we could check that with a test as well. |
edit: My original check had a bug, but the numbers are still in favour or the new distance (and hence exp/log). |
Co-authored-by: Mateusz Baran <mateuszbaran89@gmail.com>
Hm, something broke on CI unrelated to this PR I think? See https://github.com/JuliaManifolds/Manifolds.jl/actions/runs/5304573985/jobs/9600993369?pr=631 |
OrdinaryDiffEq precompilation fails, that looks unrelated to Manifolds.jl. |
Nightly checks are failing. Is it still okay to squash and merge or not? |
Yes, it's perfectly fine to squash and merge with nightly failures. I like to checked them once in a few months to see if there is something to report but that's all. |
Replaces distance, exponential, and logarithmic maps to more numerically stable versions, as per the Matlab Manopt version.