cv.doseInt: Conduct cross-validation for dose interval
In lixiaomao/personalized-dosing-interval: Determining Personalized Dosing Interval

Description Usage Arguments Examples

This function is the wrapper of cross-validation for doseInt

cv.doseInt(train, test = NULL, pred0 = NULL, nfolds = 5, type = "PDI",
  two.sided = FALSE, family = c("continuous", "ordinal", "as.ordinal"),
  method = c("rq", "svmLinear", "svmRadial", "RLT", "tree", "completeLinear",
  "completeKernel"), maxiter = 20, step = 1, trace = 0, lower = TRUE,
  K = 10, global = F, margin = 0, breaks = "quantile", aL = NULL,
  aU = NULL, Eps = NULL, Lambda = c(2^(-(0:10)), 1e-08), Cost = c(1e-05,
  0.001, 0.1, seq(1, 15, 2)), Embed.mtry = c(0.2, 0.3, 0.4, 0.5), ...)

`train`	the training data set.
`test`	the testing data set. If test is NULL, use train as test.
`pred0`	the initial prediction for training data set, if NULL, default initialization is used
`nfolds`	the number of folds used in cross-validation.
`type`	indicates whether PDI or EDI is to be calculated.
`two.sided`	indicator of whether two-sided interval is considered.
`family`	specifis the methods. 'continuous' leads to DC-algorithm,'ordinal' handles the ordinal treatments, while 'as.ordinal' cuts continuous treatment into several ordinal levels and apply the ordinal algorithm.
`method`	specified methods used for DC and ordinal approaches.
`maxiter`	maximal iterations for DC algorithm
`step`	stepsize for DC algorithm
`trace`	the level of detail to be printed: 1,2,3,4
`lower`	True or False, Whether lower boundary is considered when applied to one-sided interval
`K`	level of ordinal treatment, default 10
`global`	if global data are used for ordinal approach, default False
`margin`	how large is the margin if global=False, default 0
`breaks`	ways 'as.ordinal' cuts continuous treatment into several ordinal levels, default 'quantile'. User can also choose 'uniform' or specify customized breaks
`Eps`	parameters for the relaxation
`Lambda`	candidate penalties for the quantile regression
`Cost`	candidate costs for the support vector machine
`Embed.mtry`	candidate proportions of mtry

train1=Scenario1.continuous(2000,5,1)
test1=Scenario1.continuous(10000,5,1)

cv1=cv.doseInt(train1,test1,pred0=NULL,nfolds=2,alpha=c(0.5,0.5),type='PDI',two.sided=FALSE,
               family='continuous',
               method='svmLinear',
               maxiter=20,step=1,trace=0,lower=TRUE,Lambda = 2^(-(1:5)),Cost=c(1,7,12,19))
pred1=predict(cv1$fit,test1)

train2=Scenario2.continuous(2000,6,2)
test2=Scenario2.continuous(10000,6,2)

cv2=cv.doseInt(train2,test2,pred0=NULL,nfolds=2,alpha=c(0.5,0.5),type='PDI',two.sided=FALSE,
               family='as.ordinal',
               method='svmLinear',K=20,
               maxiter=2,step=1,trace=3,lower=TRUE,Lambda = 2^(-(1:5)),Cost=c(0.0001,0.001,0.01,0.01,0.05,0.1,0.5,1))
pred2=predict(cv2$fit,test2)


train4=Scenario4.continuous(2000,5,4)
test4=Scenario4.continuous(10000,5,4)

cv4=cv.doseInt(train4,test4,pred0=NULL,nfolds=2,alpha=c(0.5,0.5),type='PDI',two.sided=TRUE,
               family='as.ordinal',
               method='svmLinear',K=20,
              maxiter=2,step=1,trace=3,lower=TRUE,Lambda = 2^(-(1:5)),Cost=c(0.0001,0.001,0.01,0.01,0.05,0.1,0.5,1))
pred4=predict(cv4$fit,test4)