-
Notifications
You must be signed in to change notification settings - Fork 27
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
Deterministic 2 #162
Deterministic 2 #162
Conversation
… correctness of the parallel implementation
…if balance == 0 --> also use the revert from back approach
…lization over block-pairs in combination with parallelization per block-pair as well as sequential over block-pairs with parallel per block-pair (for larger balancing ranges)
…graph contraction
…ruction if non-deterministic
…ndent of the machine's number of threads (the algorithms using it don't have non-determinism concerning the distribution of elements to buckets)
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
LGTM, only some nitpicks.
Currently, using fixed vertices in deterministic mode will lead to a NonSupportedOperationException
. Remove the corresponding restriction in partitioner.cpp
. Also make sure that using another gain function in deterministic mode will not silently switch to km1. I have not found the corresponding part in the code where this is enforced. If this is already possible ignore this comment.
mt-kahypar/partition/coarsening/deterministic_multilevel_coarsener.cpp
Outdated
Show resolved
Hide resolved
mt-kahypar/partition/refinement/deterministic/deterministic_label_propagation.cpp
Outdated
Show resolved
Hide resolved
There is no silent switch to km1. There is a silent switch to the deterministic refiner, but the refiner names don't pertain to an objective function (as in sequential kahypar). |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
LGTM, only the issue with GainComputation
should be fixed before merging.
This PR contains