The implementation of ScaleAP proposed in AAAI'21. If you use our software, please cite the following paper.

Hiroaki Shiokawa,  
"Scalable Affinity Propagation for Massive Datasets,"
Proceedings of the AAAI Conference on Artificial Intelligence, 35(11), 9639-9646, May, 2021.

@article{Shiokawa2021,
  title   = {{Scalable Affinity Propagation for Massive Datasets}},
  author  = {Hiroaki Shiokawa},
  journal = {Proceedings of the AAAI Conference on Artificial Intelligence},
  volume  = {35},
  number  = {11},
  pages={9639-9646},
  month   = {5},	
  year    = {2021},
  url={https://ojs.aaai.org/index.php/AAAI/article/view/17160},
}

This software is released under the MIT license. Please read LICENSE for details.

ScaleAP requires the following softwares.

• gcc Version 9.2.0 (or later)

We have confirmed that our software works on the following environments.

• CentOS 8
• MacOS 11.4

1. Install g++ on your environment.
2. Build the softwares.

$ make clean; make

3. If you can find ScaleAP and convert, the build has been successfully completed.

The input file must include the number of objects, which is followd by a list of similarities among the objects formatted as follows:

4
-0.11
-0.12
-0.13
-0.14
-0.21
-0.22
-0.23
-0.24
-0.31
-0.32
-0.33
-0.34
-0.41
-0.42
-0.43
-0.44

By letting sim(i,j) be a similarity between objects i and j, 4 represents the number of objects. Also, -0.11, -0.12, -0.13, -0.14, -0.21, -0.22, ..., -0.44 are corresponding to sim(0,0), sim(0,1), sim(0,2), sim(0,3), sim(2,0), sim(2,1), ..., sim(3,3).

ScaleAP reads the given similarities by using the CRS format, and this requires a file conversion process. To covert the input file (sample_sim.txt) into the CRS format (sample_sim.bin), we need to run convert like as follows:

$ ./convert -i sample_graph.txt -o sample_graph.bin

convert requires two options, -i and -o, that specify names of the input file and the CRS formatted file, respectively.

Finally, we can compute Affinity Propagation for sample_sim.bin by using ScaleAP like as follows:

$ ./ScaleAP -i sample_sim.bin 

ScaleAP has the following options.

• Display the clustering result when we set <preference>=-0.22; each line indicates a pair of an object ID and a corresponding cluster ID. In the following example, objects 0 and 1 are included in a cluster ID 1.

$ ./ScaleAP -i sample_sim.bin -R -p -0.22
Object	Cluster
0	1
1	1
2	2
3	3

• Hiroaki Shiokawa, "Scalable Affinity Propagation for Massive Datasets," Proceedings of the AAAI Conference on Artificial Intelligence, 35(11), 9639-9646, May, 2021.
• http://www.kde.cs.tsukuba.ac.jp/~shion/index.html.en