Skip to content

Codes for inferring the intrinsic dimension of a real network in the geometric soft configuration/hyperbolic map model

License

Notifications You must be signed in to change notification settings

networkgeometry/detecting-dimensionality

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

6 Commits
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Detecting the ultra low dimensionality of real networks

This repository contains four folders:

  1. cyclesmap: calculations of chordless cycles in a given network.
  2. create_SD: generation of $\mathbb{S}^D$ surrogates of a given network.
  3. create_feats: calculation of chordless cycles in a set of surrogates.
  4. dimension: detection of optimal dimension of a network.

The workflow to detect the optimal dimension of a given network is:

  1. Calculation of chordless cycles of the network:
    $ ./cyclesmap/cyclesmap network network_features 
  2. Generation of surrogates of the network:
    $ ./create_SD/create_SD.sh network resolution n_poll wsize nrealizations maxD
  3. calculation of the chordless cycles of the surrogates:
    $ ./crete_feats/create_feats.sh SDnets/network SDfeats/network
  4. Detection of optimal dimension using Python 3.x (from dimension folder):
    $ python -c'import dimension; dimension.dimension(SDfeats/network,network_features,["triangles", "squares","pentagons"],maxk)'

The following is a brief description of the codes contained in each folder (more information can be found in the corresponding sh and py files).

cyclesmap

Contains the program cyclesmap to calculate the number of chordless cycles of length 3,4 and 5 of a given network.

Parameters:

  • The name of a file contaning an edge list of a network.

create_SD

Contains the script create_SD.sh to generate $\mathbb{S}^D$ surrogates of a given network.

The script requires an edgelist of the given network and a parameter setting.

Parameters:

  • network: name of the network (edgelist file must be located in RealNets folder with edge extension and features file in RealFeats with csv extension)
  • resolution: number of surrogates per dimension
  • n_poll: number of points used to infer the relation T vs. Beta
  • wsize: size of the clustering interval in which to create the surrogates
  • nrealizations: number of realizations per random Beta value (default=1)
  • maxD: the script will generate surrogates from D=1 to D=maxD

The resulting surrogates will be placed in SDnets folder. Some folders will be created during the process for calculation purposes.

create_feats

Contains the script create_feats.sh to calculate chordless cycles from a set of surrogates.

Parameters:

  • surrogates folder: name of the folder containing surrogates (they should be organized by dimension, as obtained with create_SD.sh script)
  • results folder: name of the destination folder to store the results

dimension

This folder contains the dimension.py Python library which provides functions to infer the optimal dimension of a network given a set of features (chordless cycles counts) of its surrogates. The code requires a folder with surrogate features organized by dimension, as obtained with create_feats.sh script.

The main function of this library is dimension whith parameters:

  • surrogate_set: the name of a folder with surrogate features organized by dimension, as obtained with create_feats.sh
  • network_features: the name of a file containing network features obtained with cyclesmap script
  • predictors: a set of predictors (a subset of ['triangles', 'squares','pentagons'])
  • maxk: a maximum value of k to explore (maxk)

This function and returns the optimal dimension for the given network, the value of k and the accuracy for the kNN method.


Publication

Detecting the ultra low dimensionality of real networks Pedro Almagro, Marian Boguna, M. Angeles Serrano arXiv:2110.14507, https://doi.org/10.48550/arXiv.2110.14507


Pedro Almagro, Marián Boguñá, M. Ángeles Serrano

Universitat de Barcelona | Universidad de Sevilla

September 7, 2022

Questions related to code: palmagro@us.es Questions related to calculation of cycles: marian.serrano@ub.edu

About

Codes for inferring the intrinsic dimension of a real network in the geometric soft configuration/hyperbolic map model

Resources

License

Stars

Watchers

Forks

Packages

No packages published