polyreg (version 0.0.0.9600)

xvalPoly,xvalKf,xvalDnet: Cross validation on Polynomial regression

Description

Separate the dataset to training and test set, and obtain the mean absolute error (for linear regression) or accuracy (for logistic regression).

Usage

xvalPoly(xy, maxDeg, maxInteractDeg = maxDeg, use = "lm", pcaMethod =
   FALSE, pcaPortion = 0.9, glmMethod = "all",nHoldout=10000,yCol=NULL)
xvalKf(xy, nHoldout = min(10000, round(0.2 * nrow(xy))), yCol = NULL, 
    units, activation, dropout) 
xvalDnet(x,y,hidden,output="\"sigm\"",numepochs=3,pcaMethod = FALSE, 
    pcaPortion=0.9,scaleXMat=TRUE,nHoldout=min(10000, 
    round(0.2*nrow(x))))

Arguments

xy

Data matrix or dataframe with response variable in the last column.

maxDeg

Max degree for power terms.

maxInteractDeg

Max degree of interaction terms.

use

Can be "lm" for using linear regreesion, and "glm" for using logistic regression.

trnProp

Proportion of data to be used as training set.

pcaMethod

If TRUE, use PCA.

pcaPortion

If pcaMethod is TRUE, use components up to this proportion of total variance.

glmMethod

For classification problems. If there are more than two classes, this can be "all" for All-vs-All method, or "one" for One-vs-All method.

nHoldout

Size of test test.

units

Vector specifying number of units per layer. There is one element for each hidden layer, then an element for the dummy layer that produces the predictions; specify this as NA.

activation

Vector specifying the activiation functions. One element for each hidden layer, then one function producing the final prediction, e.g. 'linear' for regression and 'softmax' for classification.

units

Vector specifying number of units per layer.

x

Numerical matrix of predictor values.

y

Numerical vector (regression case) or matrix (classification case) of response values. In the classification, if there are q classes then the matrix will have q columns, exactly one 1 per row.

Value

The return value of xvalPoly is an R vector of mean absolute error (for lm) or probability of correct classification (for glm). The i-th element of the vector is for degree i.

Details

The xvalPoly function divides the data to training and test sets, and use polyFit to generate models using training data and use polyFit.predict to generate results on test set, and compare the results. The xvalKf does the same for kerasformula neural networks package.

Examples

# NOT RUN {
y <- mtcars[,1]
data <- cbind(mtcars[,-1], y) # make y column the last column
pf1 <- xvalPoly(data,2,2,"lm",0.8,FALSE)
pf2 <- xvalPoly(data,5,3,"lm",0.8,TRUE,0.9)
# }