The nlrx package provides tools to setup and execute NetLogo simulations from R. NetLogo is a free, open-source and cross-platform modelling environment for simulating natural and social phenomena. NetLogo focusses on implementation of agent-based and spatially explicit simulation models, although system dynamics models are supported as well. NetLogo is developed and maintained at the Center for Connected Learning and Computer-Based Modeling, Northwestern University, Evanston, IL. More details on NetLogo itself are available online: NetLogo online documentation
NetLogo comes with the built-in experiment tool Behavior Space that allows to setup and execute model simulations with different settings and parameter variations and to collect model output. This experiment tool can be executed via command line in combination with an XML file that contains the experiment specifications, such as runtime, variables, output measurements, stop conditions, and more. One limitation of Behavior Space is, that it only supports full-factorial parameter designs, which may not be appropriate for complex model analyses. Furthermore, Behavior Space experiment specifications are stored within the NetLogo file and are not easily accessible from R. However, in many cases it is useful to store such specifications along with the model output and analyses results in order to enable fully reproducible model analyses.
The nlrx package utilizes the commandline functionality of Behavior Space to execute NetLogo simulations directly from R. Instead of defining experiments within NetLogo Behavior Space, experiments are defined in R using the class objects of the nlrx package. These class objects hold all the information that is needed to run these experiments remotely from R, such as path to NetLogo installation folder, path to the model file and the experiment specifications itself. nlrx provides useful helper functions to generate parameter input matrices from parameter range definitions that cover a wide range of parameter exploration approaches. By storing all relevant information on simulation experiments, including the output of the model simulations in one class object, experiments can be easily stored and shared.
In summary, the nlrx package uses a similar structure as NetLogos Behavior Space but offers more flexibility and additional tools for running reproducible complex model analyses directly from R.
In order to use the nlrx package, NetLogo (>=5.3.1) needs to be installed on the system that is used to execute model simulations (local/remote). For remote execution, NetLogo needs to be installed on remote machines as well. The nlrx package provides a utility function (download_netlogo()) that can be used to download and unzip (only unix systems) a specified NetLogo version to a local folder. For windows machines, the downloaded file needs to be executed in order to install NetLogo on the local system.
Because NetLogo is executed in a Java virtual machine, Java needs to be installed on the local/remote system as well. We recommend the Oracle Java SE Development Kit 8. While the nlrx package might work without setting the Java system path explicitly, we recommend to make sure that JAVA_HOME points to the correct Java installation of the system.
You can install the released version of nlrx from CRAN with:
And the development version from GitHub with:
General information that is needed to run NetLogo simulations remotely, such as path to the NetLogo installation folder is stored within a nl class object. Nested within this nl class are the classes experiment and simdesign. The experiment class stores all experiment specifications. After attaching a valid experiment, a simdesign class object can be attached to the nl class object, by using one of the simdesign helper functions. These helper functions create different parameter input matrices from the experiment variable definitions that can then be executed by the run_nl_one() and run_nl_all() functions. The nested design allows to store everything related to the experiment within one R object. Additionally, different simdesign helper functions can be applied to the same nl object in order to repeat the same experiment with different parameter exploration methods (simdesigns).
The “Wolf Sheep Predation” model from the NetLogo models library is used to present a basic example on how to setup and run NetLogo model simulations from R.
The nl object holds all information on the NetLogo version, a path to the NetLogo directory with the defined version, a path to the model file, and the desired memory for the java virtual machine. Depending on the operation system, paths to NetLogo and the model need to be adjusted.
library(nlrx)
# Windows default NetLogo installation path (adjust to your needs!):
netlogopath <- file.path("C:/Program Files/NetLogo 6.0.3")
modelpath <- file.path(netlogopath, "app/models/Sample Models/Biology/Wolf Sheep Predation.nlogo")
outpath <- file.path("C:/out")
# Unix default NetLogo installation path (adjust to your needs!):
netlogopath <- file.path("/home/NetLogo 6.0.3")
modelpath <- file.path(netlogopath, "app/models/Sample Models/Biology/Wolf Sheep Predation.nlogo")
outpath <- file.path("/home/out")
nl <- nl(nlversion = "6.0.3",
nlpath = netlogopath,
modelpath = modelpath,
jvmmem = 1024)The experiment object is organized in a similar fashion as NetLogo Behavior Space experiments. It contains all information that is needed to generate a simulation parameter matrix and to execute the NetLogo simulations. Details on the specific slots of the experiment class can be found in the package documentation (?experiment) and the “Advanced configuration” vignette.
nl@experiment <- experiment(expname="wolf-sheep",
outpath=outpath,
repetition=1,
tickmetrics="true",
idsetup="setup",
idgo="go",
runtime=50,
evalticks=seq(40,50),
metrics=c("count sheep", "count wolves", "count patches with [pcolor = green]"),
variables = list('initial-number-sheep' = list(min=50, max=150, qfun="qunif"),
'initial-number-wolves' = list(min=50, max=150, qfun="qunif")),
constants = list("model-version" = "\"sheep-wolves-grass\"",
"grass-regrowth-time" = 30,
"sheep-gain-from-food" = 4,
"wolf-gain-from-food" = 20,
"sheep-reproduce" = 4,
"wolf-reproduce" = 5,
"show-energy?" = "false"))While the experiment defines the variables and specifications of the model, the simulation design creates a parameter input table based on these model specifications and the chosen simulation design method. nlrx provides a bunch of different simulation designs, such as full-factorial, latin-hypercube, sobol, morris and eFast (see “Simdesign Examples” vignette for more information on simdesigns). All simdesign helper functions need a properly defined nl object with a valid experiment design. Each simdesign helper also allows to define a number of random seeds that are randomly generated and can be used to execute repeated simulations of the same parameter matrix with different random-seeds (see “Advanced configuration” vignette for more information on random-seed and repetition management). A simulation design is attached to a nl object by using one of the simdesign helper functions:
All information that is needed to run the simulations is now stored within the nl object. The run_nl_one() function allows to run one specific simulation from the siminput parameter table. The run_nl_all() function runs a loop over all simseeds and rows of the parameter input table siminput. The loops are constructed in a way that allows easy parallelisation, either locally or on remote HPC machines (see “Advanced configuration” vignette for more information on parallelisation).
nlrx provides method specific analysis functions for each simulation design. Depending on the chosen design, the function reports a tibble with aggregated results or sensitivity indices. In order to run the analyze_nl function, the simulation output has to be attached to the nl object first. The simdesign class within the nl object provides a slot for attaching output results (simoutput). An output results tibble can be attached to this slot by using the simdesign setter function setsim(nl, "simoutput"). After attaching the simulation results, these can also be written to the defined outpath of the experiment object.