jovo
Addition of a Node Labelling Term
I added the graph labelling term trace(C' P) to the rQAP formulation in http://arxiv.org/pdf/1112.5507v5.pdf. The objective function is now (where alpha is chosen between 0 and 1):
2*(1-alpha)*trace(P*A*Q'*B) + alpha * trace(C'*P)
as opposed to previously
trace(P*A*Q'*B)
I wrote some basic tests for it in code/FAQ/Tests. There is an issue with code/FAQ/lines.m now finding a quadratic search error at times. I believe this may be because the objective function has now changed but did not fully understand the function and this may not be working correctly for non integer alpha.
I also wrote a wrapper for the QAP solver so you can input adjacency matrices and fixed a small bug.
To test the new code, best to run the tests in code/FAQ/Tests.
P.S.
There are a couple of name changes I made to variables which I did before I thought I should contribute. I think they're fairly inconsequential.
this is great!
fyi, we have updated our code, in the folder 'SGM' there is a function 'SGM.m' this extends our work to incorporate seeds and initialize with the convex solution.
i wonder whether the corrections you made apply there as well? if so, maybe make a new pull request for that?
(i'm thinking of depricating the old code, since we think the newer code is better).
thanks so much!
On Tue, Aug 23, 2016 at 8:00 AM, Rob [email protected] wrote:
I added the graph labelling term trace(C' P) to the rQAP formulation in http://arxiv.org/pdf/1112.5507v5.pdf. The objective function is now (where alpha is chosen between 0 and 1):
2_(1-alpha)_trace(P_A_Q'_B) + alpha * trace(C'_P)
as opposed to previously
trace(P_A_Q'*B)
I wrote some basic tests for it in code/FAQ/Tests. There is an issue with code/FAQ/lines.m now finding a quadratic search error at times. I believe this may be because the objective function has now changed but did not fully understand the function and this may not be working correctly for non integer alpha.
I also wrote a wrapper for the QAP solver so you can input adjacency matrices and fixed a small bug.
To test the new code, best to run the tests in code/FAQ/Tests.
P.S.
There are a couple of name changes I made to variables which I did before
I thought I should contribute. I think they're fairly inconsequential.
You can view, comment on, or merge this pull request online at:
https://github.com/jovo/FastApproximateQAP/pull/3 Commit Summary
- Fixed an error in sfw when no IMAX or x0 value is entered and added a test
- Modified FAQ to work for labelled graphs and added a wrapper function for graph matching as opposed to QAP
File Changes
- A code/FAQ/Tests/Isomorphism.m https://github.com/jovo/FastApproximateQAP/pull/3/files#diff-0 (33)
- A code/FAQ/Tests/LabelledVsUnLabelled.m https://github.com/jovo/FastApproximateQAP/pull/3/files#diff-1 (57)
- M code/FAQ/fun.m https://github.com/jovo/FastApproximateQAP/pull/3/files#diff-2 (4)
- M code/FAQ/fungrad.m https://github.com/jovo/FastApproximateQAP/pull/3/files#diff-3 (29)
- A code/FAQ/graphm_sfw.m https://github.com/jovo/FastApproximateQAP/pull/3/files#diff-4 (63)
- M code/FAQ/lines.m https://github.com/jovo/FastApproximateQAP/pull/3/files#diff-5 (12)
- M code/FAQ/sfw.m https://github.com/jovo/FastApproximateQAP/pull/3/files#diff-6 (29)
Patch Links:
- https://github.com/jovo/FastApproximateQAP/pull/3.patch
- https://github.com/jovo/FastApproximateQAP/pull/3.diff
— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/jovo/FastApproximateQAP/pull/3, or mute the thread https://github.com/notifications/unsubscribe-auth/AACjchR2g7OdAtE1JxliKqZ7H7xzbRRjks5qiuDygaJpZM4Jq2LP .
the glass is all full: half water, half air. neurodata.io, jovo calendar https://calendar.google.com/calendar/embed?src=joshuav%40gmail.com&ctz=America/New_York
