Open ajasja opened 6 years ago
Even on exactly the same scoring file, the solvers produces different scorer-comparison.zip results!
Ok, so something is very wrong. Which is strange, since the solver is almost unmodified between this version and the version in the old repo.
Randomness?
LPA
On 13/12/2017 16:26, Daniel Siladji wrote:
Ok, so something is very wrong. Which is strange, since the solver is almost unmodified between this version and the version in the old repo.
— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/dancsi/DiplomaThesis/issues/14#issuecomment-351424552, or mute the thread https://github.com/notifications/unsubscribe-auth/AZN4b0qLX6el1Nf5E3UQiHgJOvzPXK3aks5s_-y7gaJpZM4Q_ArN.
-- Andrej (Andy) Brodnik
University of Primorska Department of Information Science and Technology
@andy-brodnik I thought the solver was deterministic (if I run it with the same version for the same input I get the same output).
That is the beauty of multithreaded algorithms, subtle timing issues can influence the exact result (but hopefully not its size). It is also possible that the modifications I made to the clique algorithm introduced a bug.
Well, parallel algorithms are also deterministic and non-deterministic. :-|
LPA
Daniel Siladji je 14. 12. 2017 ob 11:22 napisal:
That is the beauty of multithreaded algorithms, subtle timing issues can influence the exact result (but hopefully not its size). It is also possible that the modifications I made to the clique algorithm introduced a bug.
— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/dancsi/DiplomaThesis/issues/14#issuecomment-351669662, or mute the thread https://github.com/notifications/unsubscribe-auth/AZN4b-L1jLna34lt7eEwTYZ475vCbSWfks5tAPbmgaJpZM4Q_ArN.
-- University of Primorska Department of Information Science and Technology
Since the old solver works, this is low-priority
With the old scorer a {out-name}.current file was created containing all the set of same size. This does not happen with v3.
Exact steps to reproduce are attached in the notebooks. You only have to adjust the paths to the executables. The hepran package is not needed (although I should put it somewhere soon anyway).
Sets found with old scorer:
Sets found with the new scorer