prepareZZ {NNbenchmark}

Prepare a Dataset For All Possible Formats

Description

This function modifies a dataset to the format required by a training function: data.frame, matrix or vector (numeric), pre-normalization.

Usage

prepareZZ(Z, xdmv = "m", ydmv = "v", zdm = "d", scale = FALSE)

Arguments

Z

a matrix or a data.frame representing a dataset.

xdmv

character, either "d", "m" or "v". The prefered output format for x: data.frame, matrix, vector (numeric).

ydmv

character, either "d", "m" or "v". The prefered output format for y: data.frame, matrix, vector (numeric).

zdm

character, either "d" or "m". The prefered output format for Zxy: data.frame or matrix.

scale

logical. Scale x, y and Zxy with their respective means and standard deviations.

Value

The output is a list, usually named ZZ, with the following items:

• Zxy: the original or scaled Z in the desired format (data.frame, matrix).

• x: the original or scaled x in the desired format (data.frame, matrix, vector).

• y: the original or scaled y in the desired format (data.frame, matrix, vector).

• xory: the original x or y (as vector).

• y0: the original y (as vector).

• xm0: the mean(s) of the original x.

• ym0: the mean of the original y.

• xsd0: the standard deviation(s) of the original x.

• ysd0: the standard deviation of the original y.

• uni: the univariate (TRUE) or multivariate (FALSE) status of x (Z).

• fmla: the formula y ~ x or y ~ x1 + x2 + .. + xn where n is the number of inputs variables.

The use of attach() and detach() gives direct access to the modified values of ZZ. See the examples.

Examples

library("brnn")
library("validann")

maxit <- 200  # increase this number to get more accurate results with validann:ann
TF    <- TRUE # display the plots

### UNIVARIATE DATASET
Z     <- uGauss2
neur  <- 4

## brnn
ZZ <- prepareZZ(Z, xdmv = "m", ydmv= "v", scale = FALSE) ; ht(ZZ) 
attach(ZZ)
y_pred <- ym0 + ysd0*predict(brnn(x, y, neur))
plotNN(xory, y0, uni, TF)
lipoNN(xory, y_pred, uni, TF, lwd = 4, col = 2)
ym0 ; ysd0
detach(ZZ) ; rm(y_pred)


## validann
ZZ <- prepareZZ(Z, xdmv = "m", ydmv= "v", scale = TRUE) ; ht(ZZ)
attach(ZZ)
y_pred <- ym0 + ysd0*predict(validann::ann(x, y, neur, maxit = maxit))
lipoNN(xory, y_pred, uni, TF, lwd = 4, col = 3)
ym0 ; ysd0
detach(ZZ) ; rm(y_pred)


### UNIVARIATE DATASET + LOOP
nruns  <- 10

## brnn
ZZ <- prepareZZ(Z, xdmv = "m", ydmv= "v", scale = FALSE) ; ht(ZZ) 
attach(ZZ)
Zreg <- list() ; Zreg
for (i in 1:nruns) Zreg[[i]] <- brnn::brnn(x, y, neur) 
m      <- matrix(sapply(Zreg, function(x) x$Ed) , ncol=1) ; m
best   <- which(min(m) == m)[1] ; best
y_pred <- ym0 + ysd0*predict(Zreg[[best]])
plotNN(xory, y0, uni, TF)
lipoNN(xory, y_pred, uni, TF, lwd = 4, col = 2)
detach(ZZ) ; rm(y_pred)

## validann
ZZ <- prepareZZ(Z, xdmv = "m", ydmv= "v", scale = TRUE) ; ht(ZZ)
attach(ZZ)
Zreg <- list()
for (i in 1:nruns) Zreg[[i]] <- validann::ann(x, y, size = neur, maxit = maxit) 
m      <- matrix(sapply(Zreg, function(x) x$value), ncol=1) ; m
best   <- which(min(m) == m)[1] ; best
y_pred <- ym0 + ysd0*predict(Zreg[[best]])
lipoNN(xory, y_pred, uni, TF, lwd = 4, col = 4)
detach(ZZ) ; rm(y_pred)


### MULTIVARIATE DATASET
Z    <- mDette
neur <- 5

## brnn
ZZ <- prepareZZ(Z, xdmv = "m", ydmv= "v", scale = FALSE) ; ht(ZZ)
attach(ZZ)
y_pred <- ym0 + ysd0*predict(brnn::brnn(x, y, neur))
plotNN(xory, y0, uni, TF)
lipoNN(xory, y_pred, uni, TF, lwd = 4, col = 2)
ym0 ; ysd0
detach(ZZ) ; rm(y_pred)

## validann
ZZ <- prepareZZ(Z, xdmv = "m", ydmv= "v", scale = TRUE) ; ht(ZZ)
attach(ZZ)
y_pred <- ym0 + ysd0*predict(validann::ann(x, y, neur, maxit = maxit))
lipoNN(xory, y_pred, uni, TF, lwd = 4, col = 3)
ym0 ; ysd0
detach(ZZ) ; rm(y_pred)


### INSIDE A FUNCTION
plotds <- function(Z, xdmv = "m", ydmv = "v", scale = FALSE, neurons = 3, col = 2) {
    ZZ <- prepareZZ(Z, xdmv = xdmv, ydmv= ydmv, scale = scale) 
    attach(ZZ) ; on.exit(detach(ZZ))
    y_pred <- ym0 + ysd0*predict(brnn::brnn(x, y, neurons))
    plotNN(xory, y0, uni, TF)
    lipoNN(xory, y_pred, uni, TF, lwd = 4, col = col)
    print(ht(x))
    print(ht(y))
}
plotds(uNeuroOne, scale = FALSE, neurons = 2, col = 2)
plotds(uNeuroOne, scale = TRUE,  neurons = 3, col = 3)

plotds(mFriedman, scale = TRUE,  neurons = 5, col = 4)