MOI to MatOI #7
MOI to MatOI #7
Conversation
|
Almost the same code as https://github.com/jump-dev/SCS.jl/blob/master/src/MOI_wrapper.jl |
Codecov Report
@@ Coverage Diff @@
## master #7 +/- ##
==========================================
- Coverage 86.14% 82.91% -3.23%
==========================================
Files 3 5 +2
Lines 166 281 +115
==========================================
+ Hits 143 233 +90
- Misses 23 48 +25
Continue to review full report at Codecov.
|
|
Nice, could you adapt it to be closer to the |
we did'nt plan getting the linear case. although it shudn't be much difficult - i tried extracting the matrices given LessThan, EqualTo, GreaterThan constraints |
|
is there a specific convention on using double underscore for functions? It doesn't read that well |
yep removed them now |
|
having a hard time moving from get_conic_form to MOI.copy_to. @blegat can u share some example using MOI.copy_to on GeometricConicForm in action |
|
If I understand correctly, you need something close to function get_conic_form(::Type{T}, src::MOI.ModelLike, con_idx)
dest = ConicForm{T, ...}()
idxmap = MOI.copy_to(dest, src)
return dest
endexcept that you only need the constraints in |
|
took a lot of time but using |
src/conic_form.jl
Outdated
| trimap(i::Integer, j::Integer) = i < j ? trimap(j, i) : div((i-1)*i, 2) + j | ||
| trimapL(i::Integer, j::Integer, n::Integer) = i < j ? trimapL(j, i, n) : i + div((2n-j) * (j-1), 2) | ||
| sympackedLtoU(x, n=sympackeddim(length(x))) = _sympackedto(x, n, (i, j) -> trimapL(i, j, n), trimap) | ||
| sympackedUtoL(x, n) = _sympackedto(x, n, trimap, (i, j) -> trimapL(i, j, n)) |
There was a problem hiding this comment.
Why are these here? This is specific to SCS needing to scale off diagonal entries, do you need this too in DiffOpt?
There was a problem hiding this comment.
sorry for the late response.
yes, we need at least this part to scale diagonal entries for a PSD cone. my only source of cross checking the actual values was https://github.com/cvxgrp/diffcp, which accepts/returns matrices in SCS format
There was a problem hiding this comment.
It doesn't make much sense to do it by default for the conic form. I would suggest DiffOpt to use something like
https://github.com/jump-dev/SCS.jl/blob/c1e3a8d5469e9f289dd0b3cdd87245792d266317/src/MOI_wrapper.jl#L147
In the future, if the scaled psd cone is brought back then SCS and DiffOpt could use this without any preprocessing function.
There was a problem hiding this comment.
See jump-dev/MathOptInterface.jl#531 for the scaled psd cone
There was a problem hiding this comment.
ok got it now. will need to correct the tests accordingly!
There was a problem hiding this comment.
should I check the presence of SCS/Mosek solver and edit the matrices accordingly - so that for every solver matrix A is same
Not sure to understand what you mean
There was a problem hiding this comment.
forgot to edit the comment 😅 . saw the conic solver info at https://github.com/cvxgrp/cvxpy/tree/master/cvxpy/reductions/solvers/conic_solvers and understood this trimap thing is specific to SCS and shouldn't be a part of MatOI
There was a problem hiding this comment.
@blegat is this PR good to go? or is there anything specific to SCS that I'm missing
Change Vector to AbstracVector in template type (jump-dev#4)
My bad. Took a whole week. But generates matrices!