nlrx provides a variety of different simdesigns that generate parameter input matrices from valid experiment variable definitions. These simdesigns represent different techniques of parameter space explorations. This is not a technical guide on parameter space exploration but rather a collection of code examples on how to setup these different approaches with nlrx. Details on the sensitivity analysis methods can be found in the documentation of the sensitivity package. Details on the optimization methods can be found in the documentation of the genalg package and the GenSA package. Details on the approximate bayesian computation (ABC) methods can be found in the documentation of the EasyABC package
The following section provides valid experiment setups for all supported simdesigns using the Wolf Sheep Model from the NetLogo models library.
First we set up a nl object. We use this nl object for all simdesigns:
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 simple simdesign only uses defined constants and reports a parameter matrix with only one parameterization. To setup a simple simdesign, no variables have to be defined. The simple simdesign can be used to test a specific parameterisation of the model. It is also useful to for creating animated output of one specific parameterisation (see “Capturing Spatial NetLogo Output” vignette).
simdesign_simple is most useful, if you want to imitate pressing the go button in your model in NetLogo - just a single run, with a defined parameterisation.
nl@experiment <- experiment(expname="wolf-sheep",
outpath="C:/out/",
repetition=1,
tickmetrics="true",
idsetup="setup",
idgo="go",
idfinal=NA_character_,
idrunnum=NA_character_,
runtime=50,
evalticks=seq(40,50),
metrics=c("count sheep", "count wolves", "count patches with [pcolor = green]"),
variables = list(),
constants = list("initial-number-sheep" = 20,
"initial-number-wolves" = 20,
"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"))
nl@simdesign <- simdesign_simple(nl=nl,
nseeds=3)The distinct simdesign can be used to run distinct parameter combinations. To setup a distinct simdesign, vectors of values need to be defined for each variable. These vectors must have the same number of elements across all variables. The first simulation run consist of all 1st elements of these variable vectors; the second run uses all 2nd values, and so on.
nl@experiment <- experiment(expname="wolf-sheep",
outpath="C:/out/",
repetition=1,
tickmetrics="true",
idsetup="setup",
idgo="go",
idfinal=NA_character_,
idrunnum=NA_character_,
runtime=50,
evalticks=seq(40,50),
metrics=c("count sheep", "count wolves", "count patches with [pcolor = green]"),
variables = list('initial-number-sheep' = list(values=c(10, 20, 30, 40)),
'initial-number-wolves' = list(values=c(30, 40, 50, 60))),
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"))
nl@simdesign <- simdesign_distinct(nl=nl,
nseeds=3)The full factorial simdesign creates a full-factorial parameter matrix with all possible combinations of parameter values. To setup a full-factorial simdesign, vectors of values need to be defined for each variable. Alternatively, a sequence can be defined by setting min, max and step. However, if both (values and min, max, step) are defined, the values vector is prioritized.
nl@experiment <- experiment(expname="wolf-sheep",
outpath="C:/out/",
repetition=1,
tickmetrics="true",
idsetup="setup",
idgo="go",
idfinal=NA_character_,
idrunnum=NA_character_,
runtime=50,
evalticks=seq(40,50),
metrics=c("count sheep", "count wolves", "count patches with [pcolor = green]"),
variables = list('initial-number-sheep' = list(values=c(10, 20, 30, 40)),
'initial-number-wolves' = list(min=0, max=50, step=10)),
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"))
nl@simdesign <- simdesign_ff(nl=nl,
nseeds=3)The latin hypercube simdesign creates a Latin Hypercube sampling parameter matrix. The method can be used to generate a near-random sample of parameter values from the defined parameter distributions. More Details on Latin Hypercube Sampling can be found in McKay 1979. nlrx uses the lhs package to generate the Latin Hypercube parameter matrix. To setup a latin hypercube sampling simdesign, variable distributions need to be defined (min, max, qfun).
nl@experiment <- experiment(expname="wolf-sheep",
outpath="C:/out/",
repetition=1,
tickmetrics="true",
idsetup="setup",
idgo="go",
idfinal=NA_character_,
idrunnum=NA_character_,
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"))
nl@simdesign <- simdesign_lhs(nl=nl,
samples=100,
nseeds=3,
precision=3)Sensitivity analyses are useful to estimate the importance of model parameters and to scan the parameter space in an efficient way. nlrx uses the sensitivity package to setup sensitivity analysis parameter matrices. All supported sensitivity analysis simdesigns can be used to calculate sensitivity indices for each parameter-output combination. These indices can be calculated by using the analyze_nl() function after attaching the simulation results to the nl object. To setup sensitivity analysis simdesigns, variable distributions (min, max, qfun) need to be defined.
nl@experiment <- experiment(expname="wolf-sheep",
outpath="C:/out/",
repetition=1,
tickmetrics="true",
idsetup="setup",
idgo="go",
idfinal=NA_character_,
idrunnum=NA_character_,
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"))
nl@simdesign <- simdesign_lhs(nl=nl,
samples=100,
nseeds=3,
precision=3)
nl@simdesign <- simdesign_sobol(nl=nl,
samples=200,
sobolorder=2,
sobolnboot=20,
sobolconf=0.95,
nseeds=3,
precision=3)
nl@simdesign <- simdesign_sobol2007(nl=nl,
samples=200,
sobolnboot=20,
sobolconf=0.95,
nseeds=3,
precision=3)
nl@simdesign <- simdesign_soboljansen(nl=nl,
samples=200,
sobolnboot=20,
sobolconf=0.95,
nseeds=3,
precision=3)
nl@simdesign <- simdesign_morris(nl=nl,
morristype="oat",
morrislevels=4,
morrisr=100,
morrisgridjump=2,
nseeds=3)
nl@simdesign <- simdesign_eFast(nl=nl,
samples=100,
nseeds=3)Optimization techniques are a powerful tool to search the parameter space for specific solutions. Both approaches try to minimize a specified model output reporter by systematically (genetic algorithm, utilizing the genalg package) or randomly (simulated annealing, utilizing the genSA package) changing the model parameters within the allowed ranges. To setup optimization simdesigns, variable ranges (min, max) need to be defined. Optimization simdesigns can only be executed using the run_nl_dyn() function - instead of run_nl_all() or run_nl_one().
nl@experiment <- experiment(expname="wolf-sheep",
outpath="C:/out/",
repetition=1,
tickmetrics="true",
idsetup="setup",
idgo="go",
idfinal=NA_character_,
idrunnum=NA_character_,
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),
'initial-number-wolves' = list(min=50, max=150)),
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"))
nl@simdesign <- simdesign_GenAlg(nl=nl,
popSize = 200,
iters = 100,
evalcrit = 1,
elitism = NA,
mutationChance = NA,
nseeds = 1)
nl@simdesign <- simdesign_GenSA(nl=nl,
par=NULL,
evalcrit=1,
control=list(max.time = 600),
nseeds=1)Approximate bayesian computation (ABC) algorithms have been increasingly used for calibration of agent-based simulation models. The nlrx package provides different algorithms from the EasyABC package. These algorithms can be used by attaching the corresponding simdesigns (simdesign_ABCmcmc_Marjoram(), simdesign_ABCmcmc_Marjoram_original(), simdesign_ABCmcmc_Wegmann()).
nl@experiment <- experiment(expname="wolf-sheep",
outpath="C:/out/",
repetition=1,
tickmetrics="true",
idsetup="setup",
idgo="go",
idfinal=NA_character_,
idrunnum=NA_character_,
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),
'initial-number-wolves' = list(min=50, max=150)),
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"))
nl@simdesign <- simdesign_ABCmcmc_Marjoram(nl=nl,
summary_stat_target = c(100, 80),
n_rec = 100,
n_calibration=200,
use_seed = TRUE,
progress_bar = TRUE,
nseeds = 1)
nl@simdesign <- simdesign_ABCmcmc_Marjoram_original(nl=nl,
summary_stat_target = c(100, 80),
n_rec = 10,
use_seed = TRUE,
progress_bar = TRUE,
nseeds = 1)
nl@simdesign <- simdesign_ABCmcmc_Wegmann(nl=nl,
summary_stat_target = c(100, 80),
n_rec = 10,
n_calibration=200,
use_seed = TRUE,
progress_bar = TRUE,
nseeds = 1)