commands:

decompose the graph to 2 subgraph, and will get $mod.decomposed.graphA.out and $mod.decomposed.graphB.out:
  -train_graph <trn-file> -out_prefix <mod-prefix-> -decompose -decomp_unlab -graph_graph 
for example:
  -train_graph sample.conll08 -out_prefix sample.mod -decompose -decomp_unlab -graph_graph 
and will get sample.mod.decomposed.graphA.out and sample.mod.decomposed.graphB.out


train the subgraph, and will get the feature model: 
  -train_graph <subgraph-file> -model_graph <mod-prefix> -graph_parser [graphcjj1][graphcjj2gc]
for example:
  -train_graph sample.mod.decomposed.graphA.out -model_graph sample.mod.graphA -graph_parser graphcjj2gc
and will get sample.mod.graphA.graphcjj2gc.feat


test the graph:
  -model_graphA <modA-file> -model_graphB <modB-file> -test <test-file> -out_prefix <out-prefix> -graph_parser [graphcjj1][graphcjj2gc] -test_max_iter <testiteration> -graph_graph 
for example:
  -model_graphA sample.mod.graphA.graphcjj2gc.feat -model_graphB sample.mod.graphB.graphcjj2gc.feat -test sample.conll08 -out_prefix sample.out.conll08 -graph_parser graphcjj2gc -test_max_iter 200 -graph_graph



