Repository for NeuSO: Neural Optimizer for Subgraph Queries.

Data Format

Data graph must have the following format:

u{or 'd'} node_num edge_num v_label_num e_label_num
v v_id v_label
...
e src_id dst_id e_label
...

The first line, u represents undirected graph, while d represents directed graph. If the data graph is not a edge-labeled graph, then please set e_label_num = 1 and set all e_label = 0.

Query graph file format:

q node_num edge_num
v v_id v_label
...
e src_id dst_id e_label
...

The filtered graph file format:

f node_num edge_num
v v_id C(v_id)
...
e src_id dst_id C(src_id, dst_id)
...

CCG file have the following format:

t{or 'p'} node_num edge_num
v v_id v_card v_min_cost v_v_num v_1 v_2 ... v_v_num
...
w src_v_id dst_v_id join_v_id w_cost

The first line, t represents total explored, while p represents partially explored. Also, the ccg node_id must start from 0, continuous.

Note The v_id of graph should be started from 0 (not 1).

All these files are organized as the following directory structure:

- dataset/
    - dataset_name/
        - data_graph/
            - dataset_name.graph
        - query_graph/
            - query_graph_1.graph
            - ...
        - filter/
            - query_graph_1.filter
            - ...
        - ccg/
            - query_graph_1.ccg
            - ...

How to Exp

• Modify the config.yaml.
• Run python train.py to train the model.

Thanks

Thanks for the following repositories:

• pytorch-template for the template of PyTorch project.
• pytorch-GAT for the implementation of GAT which is the base of TriAT.