Output

SIMBA result structure will look like this:

result_simba
├── figures
└── pbg
    └── graph0
        ├── pbg_graph.txt
        ├── graph_stats.json
        ├── entity_alias.txt
        └── input
            ├── edge
            └── entity
        └── model0
            ├── config.json
            ├── training_stats.json
            ├── checkpoint_version.txt
            ├── embeddings.h5
            └── model.h5
        └── model1
            ├── config.json
            ├── training_stats.json
            ├── checkpoint_version.txt
            ├── embeddings.h5
            └── model.h5
        └── model2
            ├── config.json
            ├── training_stats.json
            ├── checkpoint_version.txt
            ├── embeddings.h5
            └── model.h5
    └── graph1
        ├── pbg_graph.txt
        ├── graph_stats.json
        ├── entity_alias.txt
        └── input
            ├── edge
            └── entity
        └── model
            ├── config.json
            ├── training_stats.json
            ├── checkpoint_version.txt
            ├── embeddings.h5
            └── model.h5

By default, all figures will be saved under result_simba/figures

The PBG training results will be stored in the directory result_simba/pbg. Inside this folder, each constructed graph will be saved in a distinct folder (by default result_simba/pbg/graph0), and each model trained on that graph will be saved into a separate folder (by default result_simba/pbg/graph0/model).

Inside each graph folder (e.g., result_simba/pbg/graph0):

  • pbg_graph.txt stores its edges on which PBG training is performed;

  • graph_stats.json stores the statistics associated with this graph;

  • entity_alias.txt keeps the mapping between the original entity IDs and their aliases.

  • input stores the extracted nodes (entities) and edges from pbg_graph.txt, which are prepared for PBG training.

  • model stores the training result of one parameter configuration. (by default model)