plot.ordEval {CORElearn}

Visualization of ordEval results

Description

The method plot visualizes the results of ordEval algorithm with an adapted box-and-whiskers plots. The method printOrdEval prints summary of the results in a text format.

Usage

    plotOrdEval(file, rndFile, ...) 
    
    ## S3 method for class 'ordEval'
plot(x, graphType=c("avBar", "attrBar", "avSlope"), ...)
    
    printOrdEval(x)

Arguments

x

The object containing results of ordEval algorithm obtained by calling ordEval. If this object is not given, it has to be constructed from files file and rndFile.

file

Name of file where evaluation results of ordEval algorithm were written to.

rndFile

Name of file where evaluation of random normalizing attributes by ordEval algorithm were written to.

graphType

The type of the graph to produce. Can be any of "avBar", "attrBar", "avSlope".

...

Other options controlling graphical output, used by specific graphical methods. See details.

Details

The output of function ordEval either returned directly or stored in files file and rndFile is read and visualized. The type of graph produced is controlled by graphType parameter:

• avBar the positive and negative reinforcement of each value of each attribute is visualized as the length of the bar. For each value also a normalizing modified box and whiskers plot is produced above it, showing the confidence interval of the same attribute value under the assumption that the attribute contains no information. If the length of the bar is outside the normalizing whiskers, this is a statistically significant indication that the value is important.

• attrBar the positive and negative reinforcement for each attribute is visualized as the length of the bar. This reinforcement is weighted sum of contributions of individual values visualized with avBar graph type.

• avSlope the positive and negative reinforcement of each value of each attribute is visualized as the slope of the line segment connecting consequent values

The avBar and avSlope produce several graphs (one for each attribute). In order to see them all on an interactive device use devAskNewPage. On some platforms or in RStudio environment the graphical window stores the history and one can browse through recent pages. Alternatively use any of non-interactive devices such as pdf or postscript. Some support for opening and handling of these devices is provided by function preparePlot. The user should take care to call dev.off after completion of the operations.

There are some additional optional parameters ... which are important to all or for some graph types.

• ciType The type of the confidence interval in "avBar" and "attrBar" graph types. Can be "two.sided", "upper", "lower", or "none". Together with ordEvalNormalizingPercentile parameter in ordEval, ciType, ciDisplay, and ciDecorate controls the type, length and display of confidence intervals for each value.

• ciDisplay The way how confidence intervals are displayed. Can be "box" or "color". The value "box" displays confidence interval as box and whiskers plot above the actual value with whiskers representing confidence percentiles. The value "color" displays only the upper limit of confidence interval, namely the value (represented with a length of the bar) beyond the confidence interval is displayed with more intensive color or shade.

• ciDecorate controls if the reinforcement factors stretching outside the confidence intervals of possible random effects are decorated by being circled with an ellipse. The default value NULL means that there are no decorations, other values are interpreted as colors in the function draw.elipse, e.g., ciDecorate="red" draws red ellipses around statisticaly significant reinforcemnets.

• equalUpDown a boolean specifying if upward and downward reinforcement of the same value are to be displayed side by side on the same level; it usually makes sense to set this parameter to TRUE when specifying a single value differences by setting variant="attrDist1" in ordEval function.

• graphTitle specifies text to incorporate into the title.

• attrIdx displays plot for a single attribute with specified index.

• xlabel label of lower horizontal axis.

• ylabLeft label of the left-hand vertical axis.

• ylabRight label of the right-hand vertical axis.

• colors a vector with four colors specifying colors of reinforcement bars for down, down_beyond, up, and up_beyond, respectively. If set to NULL this produces black and white graph with shades of gray. The colors down_beyond and up_beyond depict the confidence interval if parameter ciDisplay="color".
  The default values are colors=c("green","lightgreen","blue","lightblue").

Value

The method returns no value.

Author(s)

Marko Robnik-Sikonja

References

Marko Robnik-Sikonja, Koen Vanhoof: Evaluation of ordinal attributes at value level. Knowledge Discovery and Data Mining, 14:225-243, 2007

Marko Robnik-Sikonja, Igor Kononenko: Theoretical and Empirical Analysis of ReliefF and RReliefF. Machine Learning Journal, 53:23-69, 2003

Some of the references are available also from http://lkm.fri.uni-lj.si/rmarko/papers/

See Also

ordEval, helpCore, preparePlot, CORElearn

Examples

    # prepare a data set
    dat <- ordDataGen(200)

    # evaluate ordered features with ordEval
    oe <- ordEval(class ~ ., dat, ordEvalNoRandomNormalizers=200)
    plot(oe)
    # printOrdEval(oe)
    
    # the same effect we achieve by storing results to files
    tmp <- ordEval(class ~ ., dat, file="profiles.oe", 
                  rndFile="profiles.oer", ordEvalNoRandomNormalizers=200)   
    plotOrdEval(file="profiles.oe", rndFile="profiles.oer",
                graphType="attrBar")
    # clean up for the sake of R package checks
    file.remove("profiles.oe")
    file.remove("profiles.oer")