Closed knutwalker closed 5 years ago
Thanks a lot Paul.
@knutwalker we should see in which other algorithms we could/should use those new local random generators.
How does the performance compare between LPA on Huge vs. Heavy?
@jexp Louvain and Infomap could make use of the random generators.
I've benchmarked the performance of the plain iterators, but not yet of the complete LPA impls
@jexp seems like the HugeLPA doesn't do too well and the changes introduced a regression to the regular LPA as well
Did you flamegraph it?
Inlining with inheritance?
I've got some here lpa-flamegraphs.zip
@knutwalker I was recently looking into LPA again, for using it in the LDBC Graphalytics benchmark. LPA docs says that:
At every iteration of propagation, each node updates its label to the one that the maximum numbers of its neighbours belongs to. Ties are broken uniformly and randomly. The Graphalytics benchmark requires deterministic execution, so in case of a tie, the smallest label wins. According to the benchmark specification:
Do you think it's possible to support such an approach to achieve determinism?
We could add an option for that. Default to random option for smallest
This one includes #857, so that one should be merged first.