dynsbm-gpu

Multithread implementation of statistical clustering algorithm proposed by Matias and Miele in paper Statistical clustering of temporal networks through a dynamic stochastic block model.

Algorithm written as a part of my master thesis. It can run even 71 times faster on GPU than on single CPU thread.

Installation

Requirements

• python >= 3.5
• TensorFlow GPU >= 1.4 (nightly version is recommended due to tf.diag support on GPU)
• CUDA related libraries (follow TensorFlow installation guide)

Usage

Run methods

To run single test main.py or main_run_set.py file can be used:

• python3 main.py,
• python3 main_run_set.py data_set use_gpu init_method iteration (use_gpu = true/false, init_method = k-means/random)

There is also run_all_data_sets.sh file which run algorithm on all files in DataSet folder. On top of this script there are few variables which are used to filter out some of the data sets.

• ./run_all_data_sets.sh > results.csv.

Results

Algorithm results are printed into standard output. main.py contains descriptive (but limited) output. main_run_set.py on the other hand print results in CSV-like fashion (data is pipe separated '|').

Data sets

In this repository there is 42 data sets. Each file is named T_N_Q_K_distribution.csv, where T is number of time steps, N number of vertices, Q number of clusters, K maximum weight and distribution is name of distribution used to generate weights in graph. Algorithm itself do not require this naming convention to work, but run_all_data_sets.sh use it to filter out some of the data sets.

References

• Matias, C., & Miele, V. (2017). Statistical clustering of temporal networks through a dynamic stochastic block model. Journal of the Royal Statistical Society: Series B (Statistical Methodology), 79(4), 1119-1141. ISO 690