SCM Repository

[chemosensors] View of /pkg/man/DriftNoiseModel-class.Rd
ViewVC logotype

View of /pkg/man/DriftNoiseModel-class.Rd

Parent Directory Parent Directory | Revision Log Revision Log

Revision 55 - (download) (as text) (annotate)
Wed Mar 2 14:48:07 2016 UTC (2 years, 11 months ago) by variani
File size: 5295 byte(s)
new Rd produced by roxygen2
% Generated by roxygen2: do not edit by hand
% Please edit documentation in R/ChemosensorsClass.R, R/DriftNoiseModelClass.R, R/DriftNoiseModelClassMethods.R
\title{Method ndcomp.}








\S4method{initialize}{DriftNoiseModel}(.Object, num = "numeric",
  datasetDriftNoiseModel = "character", pck = "character",
  dsd = "numeric", ndcomp = "numeric", ndvar, dmodel = "character",
  nsd = "numeric", ...)

\item{...}{parameters of constructor.}
Character vector of model names.

List of the default parameters.
Method ndcomp.

Method ndvar.

Method dspace.

Class \code{DriftNoiseModel} generates the drift noise in a multi-variate manner
in several steps.

Function to get model names of class \code{\link{DriftNoiseModel}}.

Function to get default constructor parameters of class \code{\link{DriftNoiseModel}}.

Constructor method of DriftNoiseModel Class.

Wrapper function DriftNoiseModel.
The primary question arising in drift modeling is related to the way of one defines the drift
phenomena for gas sensor arrays. We propose to evaluate a drift subspace via common principal component analysis.
The hypothesis of common principal component analysis states that exists an orthogonal matrix \code{V}
such that the covariance matrices of \code{K} groups have the diagonal form simultaneously.
The resulted eigenvectors (columns of the matrix \code{V}) define the subspace common for the groups
and orthogonal across the components.

A preliminary step involves quantification of drift-related data 
presented in the long-term UNIMAN dataset. These results are stored
in \code{\link{UNIMANdnoise}} dataset.

On the next step the drift is injected in the sensor array data 
by generating the noise by multi-dimensional random walk 
based on the multivariate normal distribution with zero-mean and diagonal covariance matrix
- in the sub-space defined by the matrix \code{V}.
The relative proportion along the diagonal elements in the covariance matrix is specified by the importance of drift
components in terms of of projected variance.

On the final step the component correction operation 
is recalled to induce the generated noise from the random walk 
back into the complete multivariate space of the sensor array data.

Slots of the class:
  \code{num} \tab Sensor number (\code{1:17}), which drift profile is used. The default value is \code{c(1, 2)}. \cr
  \code{dsd} \tab Parameter of standard deviation used to generate the drift noise. The deault value is 0.1. \cr
  \code{ndcomp} \tab The number of components spanning the drift sub-space. The default number is 2. \cr
  \code{ndvar} \tab The importance values of drift components. The default values are \code{\link{UNIMANdnoise}} dataset. \cr
  \code{driftModel} \tab Drift model of class \code{DriftCommonModel}. \cr
Methods of the class:
  \code{predict} \tab Generates multi-variate noise injeted to an input sensor array data. \cr
  \code{dsd} \tab Gets the noise level. \cr
  \code{dsd<-} \tab Sets the noise level. \cr

The \code{plot} method has three types (parameter \code{y}):
  \code{noise} \tab (default) Depicts the drift noise generated by the model with a linechart. \cr
  \code{pc} \tab  Shows the drift components in a PCA scoreplot of an input sensor array data (parameter \code{X}.\cr
In the case \code{num} is different from value \code{c(1:17)}, 
the number of components is not the same as in \code{V} matrix.
First, the colums in \code{V} matrix are selected according to numbers pointed in \code{num}.
Second, QR-decomposition of the resulted matrix is performed to orthogonolize the component vectors.
# model: default initialization
dn <- DriftNoiseModel()

# get information about the model


# model: custom parameters
# - many sensors
dn <- DriftNoiseModel(dsd=0.5, ndcomp=3, num=1:17)



# method plot
#  - plot types 'y': barplot, noise, walk
dn <- DriftNoiseModel() # default model

plot(dn, "noise", main="plot(dn, 'noise')") 
# default plot type, i.e. 'plot(dn)' does the same plotting

X <- UNIMANshort$dat[, num(dn)]
plot(dn, "pc", X, main="plot(dn, 'pc', X)")

### example with a SensorArray
sa <- SensorArray(num = 1:5)

set <- c("A 0.01", "A 0.05", "C 0.1", "C 1")
sc <- Scenario(rep(set, 10))
conc <- getConc(sc)

sdata <- predict(sa, conc)

p1 <- plotPCA(sa, conc = conc, sdata = sdata, air = FALSE, 
  main = "feature: transient")

p2 <- plotPCA(sa, conc = conc, sdata = sdata, feature = "ss", 
  main = "feature: steady-state")

p3 <- plotPCA(sa, conc = conc, sdata = sdata, feature = "step", 
  main = "feature: step")
\code{\link{UNIMANdnoise}}, \code{\link{SensorArray}}
ViewVC Help
Powered by ViewVC 1.0.0  
Thanks to:
Vienna University of Economics and Business University of Wisconsin - Madison Powered By FusionForge