Network Analysis in R: Visualizing Network Dynamics

-
Network analysis is just a moniker for graphically describing network relationships. Whether you are a health official trying to describe the spread of communicable diseases or a business analyst describing the progress of a sales campaign or incentive, network analysis helps others view and better understand a network dynamic. You will need to download the 'network' package for this.

In this post I will be doing the following:
  • Provide a simple made up example to understand what network analysis is.
  • Expand upon the simple example by adding hyper edges, different shapes and colors, and changing labels for vertices and edges to convey additional information.
  • Provide R code with explanations of how to generate these graphics.
Let's begin with a quick example so it is clear what network analysis is.
At its simplest, a network analysis is a graphical depiction of the movement of some unit among various entities. In the above graphic, I have nine entities with the arrows or "edges" describing how they interact. This could represent the flow of goods among various businesses, or the movement of animal populations among adjacent locales. If you decided to read this post, perhaps you already have some concept in mind which you could express this way.

Here's the R code for that simple network plot.
You'll notice that there are two vectors of information containing values that describe the movement from one entity to another. The first value from from is one and from to is two. This describes a relationship ship from Entity 1 to Entity 2. You'll notice the 'directed=T' option in the call to the network function. This indicates that I want the "edges" to appear as arrows which specifies a direction to the relationship. Otherwise, there would be an ordinary line connecting the two entities. The 'vertex.sides' argument allows you define the kind of shape you want. For example, a value of 3 there would generate a triangle, 4 a square, and so forth. Large numbers generates a circle. I entered 90, but any large number would generate a circle. With that said, I'm hoping that code seems more straightforward.

Let's take our simple plot and add some additional features to it. I'm going to establish some context around this next example so it'll make more sense. Let's say that each entity is a financial institution and that the edges (arrows) represent transactions carried out by a particular person. You are a forensic accountant and want to visualize the movement of funds from this individual's various accounts. With this graphic we want to distinguish between debits and credits and also show how much money was associated with each transaction.

This means that we need to add hyper edges which are shapes that represent the units passed between entities. In this case, I'll be using triangles to represent the funds transferred between financial institutions. I'll also label those hyper-edges with the quantity associated with each transfer. To accomplish this, I used the following code.
Please note that I am still using the same network inputs as the simple example from before. All I have changed is the hyper option, added labels for the hyper-edges, and shape and color specifications for them, as well. There are 14 transactions (numbers in the from, to vectors) that took place so I added 14 additional arguments in the label, shape, and color inputs. I added a legend to this graph to describe which are credits or debits. This is the legend function in R's base plotting functionality.
The arrangement of vertices are randomly generated (not truly random, of course) so if you don't like how it looks, just rerun the plot function! You will likely end up re-plotting the graphic a number of times to get it to appear without much overlap among all the different labels and edges. There are many different ways you can control the arrangement and look of your network plot. For additional detailed information you can run vignette("networkVignette"), in the console, and it will pull up an article written by the author that goes over many of the features. Again, there are many features I haven't discussed that await you in the documentation.

Feel free to ask me any specific questions at this site. Be aware that there is a nominal $1.50 fee to submit questions. That is because it takes time and effort to respond to your questions.

I hope you found this post helpful and please leave any constructive feedback or requests a topic for a post in the comments.

Comments

Post a Comment

Popular posts from this blog

How to Get Started Playing Super Metroid / Link to the Past Crossover Randomizer.

-
I've purchased and enjoyed classic consoles like the Super Nintendo Classic and replayed some of the beloved games from my childhood. After redoing a game once, I'm good for a while, and am left hoping for some way to make it more interesting once the nostalgia wears off. One way to make classic games more interesting was to watch speed runs. I've been a fan of watching speed running classic games, like Super Metroid, for some time now. Recently, I noticed many runners starting to participate in randomizer races and was immediately fascinated. The dynamic is different than typical speed-running which focuses on runners mastering mostly repetitive strategies and precise execution. Admittedly, that is still quite entertaining, however, randomizers add a new dimension where you have to think through game logic in order progress. It also adds an element where your regular Joe can play the game and enjoy a new twist every time without spending countless hours mastering particu...
Read more

Two-Step fix for rJava library installation on Mac OS

-
I've been using R for some time and over the years I've had one consistent nagging problem.  rJava Perhaps the single most temperamental library in the whole history of R. If you are like me, you likely try to avoid anything Java based, like using openxlsx  instead of xlsx . I don't use Java but a number of libraries I do use, have it as a dependency. For example, I like to use qdap because it has a lot of nice tools for qualitative analysis, which of course uses Java. The big problem is that rJava never installs properly and gives some error along the lines of not being able to find jdk files, jni.h, or Java home when you try to call the library. I have a couple quick steps here that can get rJava up and running quickly. I haven't noticed this issue in Windows which means the library is probably written for Windows and the developer hasn't bothered to make it function out-of-the-box on Mac OS. Two quick steps and you can get rJava working in R on Mac OS. Downloa...
Read more

Structural Machine Learning in R: Predicting Probabilistic Offender Profiles using FBI's NIBRS Data

-
Image
What is Structural Machine Learning? Most machine learning tasks are designed for classifying data, but what if you have multiple outcomes? Not only do you have multiple outcomes (Y's or output variables), but you also need to enforce specific relationships among those outcomes and your predictors (X). Traditional ML can do many things, but this is not one of them. Traditional ML classifiers have one outcome and they attempt to classify that univariate outcome. A genre of machine learning that handles multiple outcomes and allows data scientists to specify a structure among all variables of interest is called "structured prediction". This genre of ML has existed in the literature for many years, but isn't something I've come across much, so I figured I would present a simple form of structured prediction and motivate it with an illustrative example. The Problem I'll motivate this with the idea of criminal profiling. It's an interesting subject and somethin...
Read more