Implementation of the method described in Local Patterns and Supergraph for Chemical Graph Classification with Convolutional Networks (S SSPR2018, pdf here).

• bin/supergraph : compute a supergraph of a graph database as well as the projections of input data onto the supergraph
• bin/supergraph-cv : compute a supergraph for each cross-validation subset
• bin/supergraph-cv-omp : the same, multi-threaded with OpenMP
• bin/project : compute the projections of the database's graphs onto a supergraph
• bin/stars : compute stars present in a graph database encoded as a vector
• bin/ds2json : convert a .ds dataset into json format
• bin/ds_stats : compute stats of a dataset
• bin/cv_indices : create files of indices for each cross-validation subset

It is possible to create the program bin/supergraph-cv-omp using a bipartite approximation of graph edit distance at each step of the algorithm by using the rule supergraph-cv-bipartite of the Makefile. This accelerates the computation but the resulting supergraph will be more dense. For more information about this, please consult the paper.

lsape is a toolbox to address the linear sum assignment problem with edition (Greyc lab).
graph_lib is toolbox of graph algorithms, especially adressing the problem of Graph Edit Distance estimation (Greyc lab).
edmonds is an implementation of the Edmonds-Blossom algorithm to find minimum weight perfect matchings on graphs.

* lsape                 https://bougleux.users.greyc.fr/lsape
* graph_lib             https://github.com/bgauzere/graph-lib
* edmonds               https://github.com/lacop/edmonds

It is then necessary to update the global variables in the Makefile :

• LSAPE_DIR Path to the repertory include of lsape
• GRAPH_DIR Path to the repertory include of graph_lib
• GRAPH_OBJ Path to the file graphlib.a
• EDMONDS_DIR Path to the root repertory of edmonds

for graph_lib dependences are :

• LSAPE
• TinyXML
• Eigen3
• OpenMP (for the multi-thread version)