R is great for anyone who wants to get started on learning Simulation. (Both Discrete Event or Agent-based, with stochastic elements in the process.)
This post is inspired by Matt Asher’s “quick-and-dirty” R simulation work on Population Growth. Matt uses it to create aRt. I felt that his core idea provided a very good framework to introduce various concepts of simulation.
The Basic Idea behind this Simulation
We take an area (it could be a country, state, city) that is made up of units (cells) that can be settled (occupied).
We allow the initial ‘pioneers’ to settle first. (Seeding the cells)
We then establish some rules for future residents to settle. (Settling Rules)
Plot the area, and collect some stats to study the growth of the population under different starting conditions and different settling rules.
Pseudo-Code for the Simulation
Print out the Plot of the Area
A Few Changes from Matt’s R code
- I switched from his matrix to a dataframe. I also switched from his image to ggplot.
- Delineated the initial parameters, seeding and settling clearly so that anyone who downloads the code can experiment
- Created a few different Seeding Functions for the pioneers (beyond random seeding)
- Matt allowed settling if the new cell was adjacent to an occupied cell. I played around with a few different population settling functions. (rectilinear, diagonal etc).
- Added a function to collect a few statistics regarding population growth
- Ran the simulation a few 100 times (without plotting) to compare different growth and settling schemes
Color Coding the Population
- Black cells : Unoccupied
- Blue cells: Pioneers
- Orange cells: Settlers who found a cell in 1-2 attempts
- Yellow cells: Settlers who found a cell after more than 2 scouting attempts
The full R code I used for these R can be found here.
Here are some runs with a 30×30 area.
Here’s what one run looked like. (Blue cells are the initial pioneers. The other colors represent the number of steps taken for them to find a home cell.)
Trying Out Different Seeding Schemes (for the pioneers)
As opposed to starting out with randomly strewn cells, what if we started out with a city downtown (a densely populated central rectangle) and let the settlers follow them?
One final seeding scheme I tried was to have two thick bands (columns of pioneer cells) and to let the population grow from there. Seeding function:
Different Population Growth Schemes
The default settling rule that Matt Asher uses is a very good one, one that I like. If a given cell is adjacent to any occupied cell (one of its 8 neighbors) then it becomes a valid cell to settle down in.
Just for experimentation, I tried a few other schemes.
1. Rectilinear Growth
Rule: You can only settle down in the 4 cells that are due N, S, East or West of an occupied cell.
2. Diagonal Growth
Rule: A settler can occupy any cell that is diagonally adjacent to another occupied cell.
Now that we have all the functionality that we wanted, it is time to start collecting some statistics for each run (‘replication’). There are many possibilities for what we could collect in a population simulation.
Here are a couple:
- How many settlers found a ‘home cell?’
- How many scouting attempts did each settler make before they found a home?
Making Multiple Replications
Iter FoundHome NumSettlers Percent
1 1 243 350 69.42857
2 2 214 350 61.14286
3 3 213 350 60.85714
4 4 258 350 73.71429
5 5 235 350 67.14286