Search parameter grid using Caret

gridsearch_caret

##sources: http://caret.r-forge.r-project.org/training.html, http://cran.r-project.org/web/packages/caret/vignettes/caret.pdf

set.seed(107) #set seed to ensure reproduction if required

## do the setup for parallel processing as per the system available (ensure "allowparallel/seed" are set up accordingly)
## for unix/ubuntu etc:
#library(doMC)
#registerDoMC(cores = 5)
##for windows:
#library(doParallel)
#registerDoParallel(cores = 8)


inTrain <- createDataPartition(y = Sonar$Class, 
                               ## the outcome data are needed
                               p = .75,
                               ## The percentage of data in the training set
                               list = FALSE)
## By default, createDataPartition does a stratified random split of the data. 

training = data[,inTrain]   ## to keep only 75% of random rows
testing = data[,-inTrain]   ## to keep the rest of the rows for testing

plsFit <- train(Class ~ .,
                data = training,
                
                method = "pls", 
                ## check "http://caret.r-forge.r-project.org/bytag.html" for all available models

                preProc = c("center", "scale")
                ## Center and scale the predictors for the training set and all future samples.
                ## Many others are available, do check the documentation
                
                ## if only one variable, might just want to search linearly
                tuneLength = 15
                
                ## if user specific grid is to be searched, it can also be specified
                #tuneGrid = data.frame(parm1 = (0:4)/4, parm2=...)
                ## OR
                #tuneGrid = expand.grid(interaction.depth = c(1, 5, 9), n.trees = (1:30)*50, shrinkage = 0.1)
                
                ## parameters related to cross validation and performance measures
                trControl = trainControl(method="repeatedCV", repeats = 3, 
                                         classProbs = TRUE,summaryFunction = twoClassSummary
                                         selectionFunction = "tolerance")
                ## worth looking into trainControl for the resampling methods and parallism options
                ## such as "seeds" and "allowparallel". Also, for choosing optimal model, "best/oneSE" available.
                ## use method = "none" if pre defined parms are to be used.
                
                ## many matrices are available to evaluate models - "RMSE" and "Rsquared" for regression and 
                ## "Accuracy" and "Kappa" for classification.for user specific, use "summaryfunction" above
                metric = ROC # Because we are using summaryfunction
                
                verbose = T ## to indicate the overall training process
                )
## Prediction:
#predict(gbmFit3, newdata = head(testing))
#predict(gbmFit3, newdata = head(testing), type = "prob")

## Comparing multiple models:
#resamps <- resamples(list(GBM = gbmFit3,
#                          SVM = svmFit,
#                          RDA = rdaFit))
#resamps
#summary(resamps)
#trellis.par.set(caretTheme())
#dotplot(resamps, metric = "ROC")
#difValues <- diff(resamps)
#difValues
#summary(difValues)

## To test the best obtained model:
# confusionMatrix(data = prediced_class, testing$Class)