GraMi is a novel framework for frequent subgraph mining in a single large graph, GraMi outperforms existing techniques by 2 orders of magnitudes. GraMi supports finding frequent subgraphs as well as frequent patterns, Compared to subgraphs, patterns offer a more powerful version of matching that captures transitive interactions between graph nodes (like friend of a friend) which are very common in modern applications. Also, GraMi supports user-defined structural and semantic constraints over the results, as well as approximate results.
For more details, check our paper: Mohammed Elseidy, Ehab Abdelhamid, Spiros Skiadopoulos, and Panos Kalnis. "GRAMI: Frequent Subgraph and Pattern Mining in a Single Large Graph. PVLDB, 7(7):517-528, 2014."
README ................... This file
LICENSE.txt .............. License file (Open Source)
build .................... build GraMi java binary files
grami .................... script to run GraMi
Datasets/ ................ Example graphs
GRAMI_*/ ................. Directory containing GraMi sources
Java JRE v1.6.0 or later
- Uncompress grami using any compression tool
- Build Java binaries using the "build" script file
- Run GraMi using "grami" script
1- Show GraMi breif help: "./grami -h"
2- Find frequent subgraphs in the "mico" undirected graph, with minimum frequency = 14000: "./grami -f mico.lg -s 14000 -t 0 -p 0"
3- Find frequent subgraphs in the "mico" undirected graph, with minimum frequency = 9340 and approximation: "./grami -f mico.lg -s 9340 -t 0 -p 0 -approxA 0.0002 -approxB=0"
4- Find frequent patterns in the "citeseer" directed graph, with minimum frequency = 160 and maximum distance bound (edge weight) = 200: "./grami -f citeseer.lg -s 160 -t 1 -p 1 -d 200"
@Ehab-abdelhamid
@ElSeidy