R: dmcFit

dmcFit {DMCfun}

R Documentation

dmcFit

Description

Fit theoretical data generated from dmcSim to observed data by minimizing the root-mean-square error ("RMSE") between a weighted combination of the CAF and CDF functions using optim (Nelder-Mead). Alternative cost functions include squared percentage error ("SPE"), and g-squared statistic ("GS").

Usage

dmcFit(
  resOb,
  nTrl = 1e+05,
  startVals = list(),
  minVals = list(),
  maxVals = list(),
  fixedFit = list(),
  freeCombined = list(),
  fitInitialGrid = TRUE,
  fitInitialGridN = 10,
  fixedGrid = list(),
  nCAF = 5,
  nDelta = 19,
  pDelta = vector(),
  tDelta = 1,
  deltaErrors = FALSE,
  spDist = 1,
  drOnset = 0,
  drDist = 0,
  drShape = 3,
  drLim = c(0.1, 0.7),
  bndsRate = 0,
  bndsSaturation = 0,
  rtMax = 5000,
  costFunction = "RMSE",
  printInputArgs = TRUE,
  printResults = FALSE,
  optimControl = list(),
  numCores = 2
)

Arguments

`resOb`	Observed data (see flankerData and simonTask for data format) and the function dmcObservedData to create the required input from either an R data frame or external .txt/.csv files
`nTrl`	Number of trials to use within dmcSim.
`startVals`	Starting values for the to-be estimated parameters. This is a list with values specified individually for amp, tau, drc, bnds, resMean, resSD, aaShape, spShape, spBias, sigm (e.g., startVals = list(amp = 20, tau = 200, drc = 0.5, bnds = 75, resMean = 300, resSD = 30, aaShape = 2, spShape = 3, spBias = 0, sigm = 4, bndsRate=0, bndsSaturation=0)).
`minVals`	Minimum values for the to-be estimated parameters. This is a list with values specified individually for amp, tau, drc, bnds, resMean, resSD, aaShape, spShape, spBias, sigm (e.g., minVals = list(amp = 0, tau = 5, drc = 0.1, bnds = 20, bndsRate=0, bndsSaturation=0, resMean = 200, resSD = 5, aaShape = 1, spShape = 2, spBias = -20, sigm = 1)).
`maxVals`	Maximum values for the to-be estimated parameters. This is a list with values specified individually for amp, tau, drc, bnds, resMean, resSD, aaShape, spShape, spBias, sigm (e.g., maxVals = list(amp = 40, tau = 300, drc = 1.0, bnds = 150, bndsRate=1, bndsSaturation=500, resMean = 800, resSD = 100, aaShape = 3, spShape = 4, spBias = 20, sigm = 10))
`fixedFit`	Fix parameter to starting value. This is a list with bool values specified individually for amp, tau, drc, bnds, resMean, resSD, aaShape, spShape, spBias, sigm (e.g., fixedFit = list(amp = F, tau = F, drc = F, bnds = F, bndsRate=T, bndsSaturation=T, resMean = F, resSD = F, aaShape = F, spShape = F, spBias = T, sigm = T)) NB. Value if fixed at startVals.
`freeCombined`	If fitting 2+ datasets at once, which parameters are allowed to vary between both fits (default = all parameters fixed between the two fits e.g. parameter = F). This is a list with bool values specified individually for amp, tau, drc, bnds, resMean, resSD, aaShape, spShape, spBias, sigm (e.g., freeCombined = list(amp = F, tau = F, drc = F, bnds = F, bndsRate=F, bndsSaturation=F, resMean = F, resSD = F, aaShape = F, spShape = F, spBias = F, sigm = F))
`fitInitialGrid`	TRUE/FALSE
`fitInitialGridN`	10 linear steps between parameters min/max values (reduce if searching more than ~2/3 initial parameters)
`fixedGrid`	Fix parameter for initial grid search. This is a list with bool values specified individually for amp, tau, drc, bnds, resMean, resSD, aaShape, spShape, spBias, sigm (e.g., fixedGrid = list(amp = T, tau = F, drc = T, bnds = T, bndsRate=T, bndsSaturation=T, resMean = T, resSD = T, aaShape = T, spShape = T, spBias = T, sigm = T)). As a default, the initial gridsearch only searches the tau space.
`nCAF`	The number of CAF bins.
`nDelta`	The number of delta bins.
`pDelta`	An alternative option to nDelta (tDelta = 1 only) by directly specifying required percentile values (vector of values 0-100)
`tDelta`	The type of delta calculation (1=direct percentiles points, 2=percentile bounds (tile) averaging)
`deltaErrors`	TRUE/FALSE Calculate delta bins for error trials
`spDist`	The starting point (sp) distribution (0 = constant, 1 = beta, 2 = uniform)
`drOnset`	The starting point of controlled drift rate (i.e., "target" information) relative to automatic ("distractor" incormation) (> 0 ms)
`drDist`	The drift rate (dr) distribution type (0 = constant, 1 = beta, 2 = uniform)
`drShape`	The drift rate (dr) shape parameter
`drLim`	The drift rate (dr) range
`bndsRate`	Collapsing bounds rate. 0 (default) = fixed bounds
`bndsSaturation`	Collapsing bounds saturation
`rtMax`	The limit on simulated RT (decision + non-decisional components)
`costFunction`	The cost function to minimise: root mean square error ("RMSE": default), squared percentage error ("SPE"), or likelihood-ratio chi-square statistic ("GS")
`printInputArgs`	TRUE (default) /FALSE
`printResults`	TRUE/FALSE (default)
`optimControl`	Additional control parameters passed to optim (see optim details section)
`numCores`	Number of cores to use

Value

dmcfit returns an object of class "dmcfit" with the following components:

`sim`	Individual trial data points (RTs for all trial types e.g., correct/error trials) and activation vectors from the simulation
`summary`	Condition means for reaction time and error rate
`caf`	Conditional Accuracy Function (CAF) data per bin
`delta`	DataFrame with distributional delta analysis data correct trials (Bin, meanComp, meanIncomp, meanBin, meanEffect)
`delta_errs`	DataFrame with distributional delta analysis data incorrect trials (Bin, meanComp, meanIncomp, meanBin, meanEffect)
`par`	The fitted model parameters + final cost value of the fit

Examples


# Code below can exceed CRAN check time limit, hence donttest
# Example 1: Flanker data from Ulrich et al. (2015)
fit <- dmcFit(flankerData) # only initial search tau
plot(fit, flankerData)
summary(fit)

# Example 2: Simon data from Ulrich et al. (2015)
fit <- dmcFit(simonData) # only initial search tau
plot(fit, simonData)
summary(fit)

# Example 3: Flanker data from Ulrich et al. (2015) with non-default
# start vals and some fixed values
fit <- dmcFit(flankerData,
  startVals = list(drc = 0.6, aaShape = 2.5),
  fixedFit = list(drc = TRUE, aaShape = TRUE)
)

# Example 4: Simulated Data (+ve going delta function)
dat <- createDF(nSubjects = 20, nTrl = 500, design = list("Comp" = c("comp", "incomp")))
dat <- addDataDF(dat,
  RT = list(
    "Comp_comp" = c(510, 100, 100),
    "Comp_incomp" = c(540, 130, 85)
  ),
  Error = list(
    "Comp_comp" = c(4, 3, 2, 1, 1),
    "Comp_incomp" = c(20, 4, 3, 1, 1)
  )
)
datOb <- dmcObservedData(dat, columns = c("Subject", "Comp", "RT", "Error"))
plot(datOb)
fit <- dmcFit(datOb, nTrl = 5000)
plot(fit, datOb)
summary(fit)

# Example 5: Fitting 2+ datasets within all common parameters values
fit <- dmcFit(list(flankerData, simonData), nTrl=1000)
plot(fit[[1]], flankerData)
plot(fit[[2]], simonData)
summary(fit)

# Example 6: Fitting 2+ datasets within some parameters values varying
fit <- dmcFit(list(flankerData, simonData), freeCombined=list(amp=TRUE, tau=TRUE), nTrl=1000)
summary(fit) # NB. amp/tau values different, other parameter values equal

[Package DMCfun version 3.5.4 Index]