Closed aldma closed 2 years ago
Merging #57 (19a0030) into master (de2260e) will increase coverage by
0.73%
. The diff coverage is93.82%
.:exclamation: Current head 19a0030 differs from pull request most recent head 765f12d. Consider uploading reports for the commit 765f12d to get more accurate results
@@ Coverage Diff @@
## master #57 +/- ##
==========================================
+ Coverage 89.03% 89.76% +0.73%
==========================================
Files 20 21 +1
Lines 857 938 +81
==========================================
+ Hits 763 842 +79
- Misses 94 96 +2
Impacted Files | Coverage Δ | |
---|---|---|
src/algorithms/panocplus.jl | 93.82% <93.82%> (ø) |
|
src/utilities/iteration_tools.jl | 84.00% <0.00%> (+1.33%) |
:arrow_up: |
src/algorithms/panoc.jl | 97.82% <0.00%> (+2.17%) |
:arrow_up: |
Continue to review full report at Codecov.
Legend - Click here to learn more
Δ = absolute <relative> (impact)
,ø = not affected
,? = missing data
Powered by Codecov. Last update de2260e...765f12d. Read the comment docs.
If I had one wish it would be for an equivalence test between NOLIP and PANOC
That would be very useful to have in case anything changes in the implementation of either algorithm, it could help catch any mistake early
I've added an equivalence test between NOLIP and PANOC and then changed the name from NOLIP to PANOCplus. Thanks for your hints!
Hello! First of all, thanks for all the work you put into this repo.
@AndThem and I would like to contribute with a variant of PANOC we recently investigated (arXiv:2112:13000).
The algorithm consists of two entangled backtracking steps: before a tentative update is accepted along some search direction, there is a check on the forward-backward stepsize
gamma
that should satisfy a Lipschitz bound. In contrast with the original PANOC, this refined linesearch procedure allows to easily reject poor steps and to handle smooth termsf
that have only locally Lipschitz continuous gradient.In the preprint we denoted this variant by PANOC+. However, to avoid issues with symbols and names, the algorithm has been implemented as a solver called NOLIP. This PR is to add NOLIP to the repo.