dreambydream / pruned-landmark-labeling

Fast and exact shortest-path distance querying index

Pruned Landmark Labeling

Pruned landmark labeling is a new shortest-path distance querying algorithm for real-world graphs, such as social networks, web graphs, biological networks and computer networks.

Advantages

The algorithm has the following advantages (for details, please see our paper):

• Fast --- it answers distance queries in microseconds,
• Scalable --- it can be applied to networks with hundreds of millions of edges,
• Exact --- unlike approximate methods, it always answers exactly correct distance, and
• Almost Parameter Free --- unlike other state-of-the-art methods, it does not require any parameter tuning.

Moreover, this implementation is:

• Easy to Use --- by copying only one header file to your project, you can start using the index.

Usage

Given a graph, it first constructs an index. Then, using the index, it can quickly answer distance between two vertices.

From CUI Interface

$ make
$ bin/construct_index samples/graph_example.tsv index_file
$ bin/query_distance index_file <<< "1 4"
2

• Execute make to build programs.
• Execute bin/construct_index to construct an index from a graph.
• Execute bin/query_distance and write pairs of vertices to STDIN to query distance between pairs of vertices.

From Your Program

PrunedLandmarkLabeling<> pll;
pll.ConstructIndex(edge_list);
cout << pll.QueryDistance(1, 4) << endl;

• Call ConstructIndex to construct an index from a graph (an edge list or a file).
• Call QueryDistance to query distance between two vertices.
• Call LoadIndex and StoreIndex to load and store an index.

For further information, please see pruned_landmark_labeling.h, samples and tests.

Details

• In a graph file, each line should contain two integers describing an edge (see samples/graph_example.tsv).
• Vertices should be described by integers starting from zero.
• Program bin/query_distance reads pairs of vertices until EOF, thus you can use it to process multiple pairs of vertices at once.
• Execute make test to run tests (google-gtest is required).

References

• Takuya Akiba, Yoichi Iwata, and Yuichi Yoshida, Fast Exact Shortest-Path Distance Queries on Large Networks by Pruned Landmark Labeling. In SIGMOD 2013, to appear.