Graph partition algo with Neo4j graph database -


i know there has famous graph partition algo tools metis implemented karypis lab (http://glaros.dtc.umn.edu/gkhome/metis/metis/overview)

but wanna know there method partition graph stored in neo4j? or have dump neo4j's data , transform node , edge format manually fit metis input format?

regarding new-ish , interesting algorithms, no means exhaustive or state of art, these first places look:

specific algorithm: didic (distributed diffusive clustering) - used once in thesis (partitioning graph databases)

  • you iterate on nodes, each node retrieve neighbors, in order spread of "some unit" neighbors
  • easy implement.
  • can made deterministic
  • iterative - it's based on iterations (like super steps in pregel) can stop @ time. longer leave better result, in theory (though in cases, on graph shapes can unstable)
  • when implemented ran 100 iterations on machine ~30gb ram, ~4 million nodes - took no more 2 days complete.

specific algorithm: evocut "finding sparse cuts locally using evolving sets" - local probabilistic algorithm microsoft - related these papers

  • difficult implement
  • local algorithm - bfs-like access patterns (random walks)
  • it's been while since read paper, remember built on clean abstractions:
    • evonibble (pluggable - decides how of neighborhood add current cluster
    • evocut (calls evonibble multiple times find local cluster)
    • evopartition (calls evocut repeatedly partition entire graph)
  • not deterministic

general algorithm family: hierarchical graph clustering

from high level:

  • coarsen graph collapsing nodes aggregate nodes
    • coarsening strategy selectable
  • find clusters in coarsened/smaller graph
    • clustering strategy selectable
  • incrementally decoarsen graph, refining @ clustering @ each step
    • refining strategy selectable

notes:

  • if graph changes (or results don't need right date) may possible coarsen once (or infrequently) work coarsened graph - save computation
  • i don't know of specific algorithm recommend

general limitations - things few clustering algorithms do:

  • node types not acknowledged - i.e., nodes treated equally
  • relationship types not acknowledged - i.e., relationships treated equally
  • relationship direction not acknowledged - i.e., relationships treated undirected

Comments

Post a Comment

Popular posts from this blog

image - ClassNotFoundException when add a prebuilt apk into system.img in android -

i'm in trouble add prebuilt apk system.img based taiwan mtk android phone platform, when push apk onto device,after reboot phone,the apk can installed succecessfully,but if add apk alps/out/target/product/$prj/system/app or vendor/mediatek/$prj/artifacts/out/target/product/$prj/system/app ,and produce system.img,the apk can't installed normally,the logcat say: a.lang.runtimeexception: unable instantiate activity componentinfo{com.sec.android.widgetapp.ap.hero.accuweather.widget/com.sec.android.widgetapp.ap.hero.accuweather.widget.weatherclock}: java.lang.classnotfoundexception: com.sec.android.widgetapp.ap.hero.accuweather.widget.weatherclock. can tell me what's problem ? when package manager installs apps part of system partition different installing apps on data partition. looks app using libraries somewhere else. when install on data lib directory created in apps data directory , symlinked /data/app-lib/. when apps installed on system it's expecte...
Read more

I need to import mysql 5.1 to 5.5? -

i have mysql 5.1 database , need import localhost running 5.5. how import 5.5 or downgrade mysql? install phpmyadmin 5.1 os: http://www.wikihow.com/install-phpmyadmin-on-your-windows-pc or https://www.digitalocean.com/community/articles/how-to-install-and-secure-phpmyadmin-on-ubuntu-12-04 then login, click on database, hit export, , export in sql format (quick fine). import .sql file 5.5 server (after installing phpmyadmin) logging in clicking database , hit import. hope helps!
Read more

Java, Hibernate, MySQL - store UTC date-time -

i saw here on few related questions (like this one , of course, this one )... essentially, want store date-time utc, , let application user choose time zone wants display date-time in. since seems date-time fields affected underlying jdbc driver, wonder if acceptable way go storing utc date-time: set both mysql , application server machine utc time zone (no need separate) both mysql , jvm should pick underlying system time settings (if not instructed otherwise) use datetime table columns on mysql side use java.util.date corresponding mapping on hibernate side (i guess java.sql.timestamp used too) let application worry interpreting date-time fields - i.e. let user choose preferred time zone is ok? edit clarify - here meant refer timestamps created strictly on server (e.g.date-time of record creation). application server instantiates date objects (new date() equals current date-time on server, , time zone agnostic). if client user wants supply date searching/filte...
Read more