optimize_SCCANsparseness: Optimization of SCCAN sparseness

In neuroconductor-releases/LESYMAP: Lesion to Symptom Mapping in R

Description

Function used to optimize SCCAN sparseness for lesion to symptom mapping.

Usage

optimize_SCCANsparseness(lesmat, behavior, mask, nFolds = 4,
  sparsenessPenalty = 0.03, lowerSparseness = -0.9,
  upperSparseness = 0.9, tol = 0.03, justValidate = FALSE,
  cvRepetitions = ifelse(length(behavior) <= 30, 6,
  ifelse(length(behavior) <= 40, 5, ifelse(length(behavior) <= 50, 4, 3))),
  showInfo = TRUE, directionalSCCAN = TRUE, mycoption = 1,
  robust = 1, sparseness = NA, nvecs = 1, cthresh = 150,
  its = 30, npermsSCCAN = 0, smooth = 0.4,
  sparseness.behav = -0.99, maxBased = FALSE, ...)

Arguments

lesmat	lesion matrix
behavior	behavior vector
mask	antsImage mask
nFolds	how many folds to use
sparsenessPenalty	penalty term
lowerSparseness	minimum searched sparseness
upperSparseness	maximum searched sparseness
tol	tolerance value, see optimize() in R
justValidate	just check the CV of provided sparseness
cvRepetitions	number of cross-validations at each sparseness value. Dynamically set depending on sample size: <=30 to 6 reps, <=40 to 5 reps, <=50 to 4 reps, > 50 to 3 reps.
showInfo	logical (default=TRUE) display messages
directionalSCCAN	(default=TRUE) switching to FALSE will switch sparseness range in the positive side, 0.005 to 0.9
mycoption	standard SCCAN parameter
robust	standard SCCAN parameter
sparseness	standard SCCAN parameter
nvecs	standard SCCAN parameter
cthresh	standard SCCAN parameter
its	standard SCCAN parameter
npermsSCCAN	SCCAN permutations
smooth	standard SCCAN parameter
sparseness.behav	what sparsness to use for behavior
maxBased	standard SCCAN parameter
...	other arguments received from lesymap or lsm_sccan.

Value

List with:
minimum - best sparseness value
objective - minimum value of objective function
CVcorrelation - cross-validated correlation of optimal sparness

Author(s)

Dorian Pustina

the optimization function Will run SCCAN on each training fold, compute behavior prediction on the test fold, and finally return a cross validated correlation from entire sample

end of optimfun

neuroconductor-releases/LESYMAP documentation built on Nov. 13, 2020, 11:28 p.m.