Gowers-Method

Gowers Method for finding latent networks of multi-modal data

Overview

This project contains the code for Gower's Method, which is a method for finding latent networks of multi-modal data. In this context, multi-modal or multi-view data, is data which has more than one mode of measurement. As a result, we employ Gower's Coefficient of Similarity combined with weighting by network entropy to find latent positions of points, which we then learn a 'best-fit' graph to those points.

Options available in this package include:

Use it on multi-modal or one-mode data.
Graph learning methods are: modularity k-NN (default), and Weighted Consensus Graph
Weighting schemes are: network entropy (default), none, and user-supplied

Installation

To install the package, you can use pip's built-in functionality with:

pip install git+https://github.com/ijcruic/Gowers-Method#egg=Gowers_Method

Usage

The general usage of the code follows the format of:

importing the package
creating a latent_graph object
adding data to that latent graph object
And, finally, fitting a graph to the data in the latent graph object

The following code details an example

from Gowers_Method import latent_graph

mode_files = ['Mode_1.csv', 'Mode_2.csv', 'Mode_3.csv']

if __name__ == "__main__":
    lg = latent_graph()
    lg.load_data_from_file(mode_files)
    network = lg.learn_graph()

Input

There are some inputs to the method to be aware of, including the data and the weighting scheme. Data can either be submitted as a list of files, where each file is a mode of the data or a list of Pandas data frames, where each data frame is a mode of the data. The weighting scheme can be unspecified, in which case it will be network entropy. It can also be specified as 'unweighted' in which case it will be unweighted (the weight vector will be all ones). If you wish to have a user specified weight scheme, you must a pass a list of length number of variables, where each entry is the numerical weight desired for that variable. Finally, graph construction can be unspecified, in modularity k-NN will be used, or specified as 'WCG' to learn a Weighted Consensus Graph.

Output

Output of the method will be the graph adjacency in a Pandas data frame, where the index and columns are the data names.

References

Campedelli G.M., Cruickshank I., Carley K.M. (2019) Detecting Latent Terrorist Communities Testing a Gower’s Similarity-Based Clustering Algorithm for Multi-partite Networks. In: Aiello L., Cherifi C., Cherifi H., Lambiotte R., Lió P., Rocha L. (eds) Complex Networks and Their Applications VII. COMPLEX NETWORKS 2018. Studies in Computational Intelligence, vol 812. Springer, Cham

ijcruic / Gowers-Method