onerva-korhonen / climate-companions-bipartite Goto Github PK

Code for investigating the even co-participation based bipartite graph structure in the Ilmastokumppanit (Climate Companions) city-to-business network. Should be applicable also to other bipartite module analyses.

Python 39.08% Jupyter Notebook 60.92%

climate-companions-bipartite's Introduction

climate-companions-bipartite

Code for investigating the event co-participation based bipartite graph structure in the Ilmastokumppanit (Climate Companions) city-to-business network. Should be applicable also to other bipartite module analyses.

functions.py contains functions needed for creating and analysing the bipartite network. The functions can be called from either a frontend script (frontend.py was used for creating the results of the related article) or interactively. parameters.py stores all parameters used in the analysis.

For a short(-ish) walkthrough of the code and main results, have a look at walkthrough.py (related data is stored at walkthrough_data). The easiest way to do that is to launch Binder (that saves you the trouble of setting package versions or even installing python):

(Preparing the environment may take a while. When it opens, just click walkthrough.py to launch the notebook).

The versions used are Python 2.7, pandas 0.22.0, Numpy 1.14.0, NetworkX 2.1, Scipy 1.0.0, and Matplotlib 2.1.2.

This work is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License.

climate-companions-bipartite's People

Contributors

Stargazers

Watchers

climate-companions-bipartite's Issues

add a separator parameter to readNodes

csv files can use different separators; better to give the separator as a parameter to the readNodes function so there's no need to edit the files
NOTE: this is now hardcoded to ;

Remove line 618 from functions

years and link paths

At the moment, years and input paths are given as separate parameters. This may cause mismatch (match between these two lists must be manually checked). So, it might be better to give just years and construct paths from them.

Add a line showing the 50% level of the index in the index-degree scatter

The % value could be defined in params and passed as a parameter

link weight in visualization

Set weight to reflect its "weight" parameter (= number of people participating)

colormap

Check the colormap (defined in parameters.py)! It should be as distinctive as possible with at least 20 clearly different values. 'tab20' would be ideal but for some reason I don't get it working - maybe it has been removed from some versions of matplotlib (or not added to others).

Definition of clique homogeneity

Define a measure of homogeneity of cliques and add the definition to the code.

Implement a function for calculating the number of non-member nodes

Update tags in params to correspond to the new data

Or could this possibly be done automatically somehow? Giving the tags manually sounds like a potential source of problems...

visualization settings function

Write a function for getting the visualization settings used in the paper (this could be used in the example frontend script). The function should give cfg as output.

Normalized diversity

Divide the effective diversity by richness to see how far away the clique is from its theoretical maximum diversity and plot this against clique size.

layout

In order to get the same layout for all network visualizations from different functions, consider calculating the layout in frontend and passing it as a parameter.

Add a "non-verbal" option for clique detection?

This would improve the readability of the notebook since there would be less output

Check how 0-degree nodes are handled

These nodes are pruned away from visualizations, but are they (and should they be) included in degree distributions etc.?

Statistical testing + Bonferroni for the mean class degrees vs random networks

networkx version

networkx 1.X and 2.X handle bipartites differently (they seem to be a separate class in nx 2.X). So, consider setting a standard version of nx in the beginning of the functions file.

testing against random: add a t-test

The significance of the difference between the observed value and the mean of random distribution could be tested as a one-sample t-test for average: is the average of random networks different from the observed value?

Remove non-member nodes from starness analysis

These nodes are companies/instances that visit some events but are not members. So, their presence most probably artificially increases starness and they should therefore be excluded from the analysis.

Check how often a star contains only one representative of a field

degree distributions: change visualizations to histograms

should the whole thing be updated to Python 3?

rename companyInputPath as participantInputPath?

This would make sense but should only needed if there's extra time

node shape in visualization

Add different node shapes for "support members" (universities) and non-member participants. To do this, nodes should at the network construction step be assigned with a membership class. Then, shapes can be set in the drawNetwork function.

Change all visualization functions to return if save path is not given

This is required for reasonably running the example notebook.