stepEvolution {DiceEval} R Documentation

## Evolution of the stepwise model

### Description

Graphical representation of the selected terms using stepwise procedure for different values of the penalty parameter.

### Usage

stepEvolution(X,Y,formula,P=1:7,K=10,test=NULL,graphic=TRUE)


### Arguments

 X a data.frame containing the design of experiments Y a vector containing the response variable formula a formula for the initial model P a vector containing different values of the penalty parameter for which a stepwise selected model is fitted K the number of folds for the cross-validation procedure test an additional data set on which the prediction criteria are evaluated (default corresponds to no test data set) graphic if TRUE the values of the criteria are represented

### Value

a list with the different criteria for different values of the penalty parameter. This list contains:

 penalty the values for the penalty parameter m size m of the selected model for each value in P R2 the value of the R2 criterion for each model

According to the value of the test argument, other criteria are calculated:

 a. If a test set is available, R2test contains the value of the R2 criterion on the test set b. If no test set is available, the Q2 and the RMSE computed by cross-validation are done.

### Note

Plots are also available. A tabular represents the selected terms for each value in P.

The evolution of the R2 criterion, the evolution of the size m of the selected model and criteria on the test set or by K-folds cross-validation are represented.

These graphical tools can be used to select the best value for the penalty parameter.

### Author(s)

D. Dupuy

step procedure for linear models.

### Examples

## Not run:
data(dataIRSN5D)
design <- dataIRSN5D[,1:5]
Y	   <- dataIRSN5D[,6]
out	   <- stepEvolution(design,Y,formulaLm(design,Y),P=c(1,2,5,10,20,30))

## End(Not run)


[Package DiceEval version 1.5 Index]