3 Notes of R and Python | Notes for Social Network Science Projects using R

3.1 Connecting R and Python for Social Network Science Projects

The following script is my notes for Social Network Science Projects to combine R and Python. Also, in my particular case, I am using a Mac OS/Linux operation system. In Windows, please launch the “Anaconda prompt”.

In the following, I will only present Mac OS/Linux operation system.

3.2 Conda environment

To begin, I will first create an environment that would contain all the python modules of the entire Social Network Science Project using conda. After installing conda in my computer, I open my terminal and add the following line of codes: conda create --name sns python=3.8 anaconda that would create a new environment called sns (Social Network Science) and would also install python version 3.8.

3.3 Installing Social Network Science modules from Python

If you are willing to activate the environment from the terminal, then you would have to use conda activate sns. Notice that I give the name sns that could be modified for whatever you prefer. We can install some modules from python activating the environment from the terminal (i.e. conda activate sns), and then running the following codes

python -m pip install numpy
python -m pip install panda
python -m pip install igraph
python -m pip install matplotlib
python -m pip install leidenalg
python -m pip install networkx
python -m pip install pyintergraph

Or, all at once - python -m pip install numpy panda igraph matplotlib leidenalg networkx pyintergraph

Check the modules installed in your environment with pip list.

In my personal computer, I had some issues with the visualizations of igraph. Hence, I install some extra modules:

python -m pip install cairocffi
python -m pip install pycairo

Also, I install cairo using the following code suggested in the webpage of pycairo, that use Homebrew:

brew install cairo pkg-config

I will also install graph-tool using brew (other options in the webpage):

brew install graph-tool

After installing some modules, you should now deactivate the session, and you would have to put the following line of code: conda deactivate in your terminal. If you are willing to check your environments, just press: conda info --envs in your terminal. If for some reason you prefer to remove your entire environment, you would have to use the following code: conda remove --name sns --all

Is often common for python users to use the Jupyter Notebook, that might be a better option if you are just going to use python. In which case, after activating your conda environment, you could set your working directory through cd [DIRECTORY] in the terminal. Then launch the Jupyter notebook by jupyter notebook.

3.4 Connecting R and Python

We can run Python codes directly from R using RMarkdown. I prefer connecting the best of both worlds. Mostly, because in my personal workflow I tend to use statistical models for social networks analysis. Some of these other statistical models that use social networks are:

Different models available in Statnet (e.g. exponential random graph models, epidemiological models, relational event models, or the latent position and cluster models for statistical networks)
Stochastic actor-oriented model
Dynamic Network Actor-Oriented Model

Now that I have an environment called sns, I could now start using R and Python together. We would need to install the reticulate package.

#install.packages("reticulate")
library(reticulate)

use_condaenv(condaenv = "sns", conda = "auto", required = TRUE)
main <- import_main()

We can also install a package from R using anaconda of Python, and run any Python code in R using py_run_string

#reticulate::conda_install(c("leidenalg", "igraph"), 
#                            envname = "sns", pip = TRUE)

py_run_string("import numpy as np")
py_run_string("my_python_array = np.array([2,4,6,8])")
py_run_string("print(my_python_array)")

Also, we can use python directly from Rmarkdown adding the chunk {python} instead of {r} in the Rmarkdown. For example, we can replicate the classic model of Holland, Laskey & Leinhardt (1983) on Stochastic Block Model using stochastic_block_model

import networkx as nx
sizes = [75, 75, 300]
probs = [[0.25, 0.05, 0.02],
        [0.05, 0.35, 0.07],
        [0.02, 0.07, 0.40]]
        
g = nx.stochastic_block_model(sizes, probs, seed=0)
len(g)

H = nx.quotient_graph(g, g.graph['partition'], relabel=True)

for v in H.nodes(data=True):
      print(round(v[1]['density'], 3))
      
for v in H.edges(data=True):
      print(round(1.0 * v[2]['weight'] / (sizes[v[0]] * sizes[v[1]]), 3))

There are neat packages out there in Python, some of my favourites are:

python: general purpose network analysis
NetworkX: general purpose network analysis
pyintergraph: convert Python-Graph-Objects between networkx, python-igraph and graph-tools
snap: general purpose network analysis
metaknowledge: computational research in bibliometrics, scientometrics, and network analysis
pySciSci: computational research in bibliometrics, scientometrics, and network analysis
leiden_lag: for community detection, with a special focus in the leiden algorithm. Some nice codes are available in the official GitHub of CWTS
graph-tool: for stochastic block models and other general purpose network analysis. Check also the databases available in Netzschleuder.
EstimNetDirected: Exponential random graph models for big networks
ALAAMEE: Autologistic Actor Attribute Model (ALAAM) parameter estimation using Equilibrium Expectation (EE) algorithm