graphics: Graphics function to plot precision/recall or f.score/recall...

In PerfMeas: Performance Measures for Ranking and Classification Tasks

Graphics function to plot precision/recall or f.score/recall curves

Description

Function to plot multiple precision/recall or f.score/recall curves

Usage

precision.recall.curves.plot(y, range=seq(from=0, to=1, by=0.1), 
curve.names=1:length(y), cex.val=0.6, f="", height=9, width=11,
col=c("black","red1","blue1","green1","darkgrey","brown1","yellow1","orange1",
"red4","blue4","green4","lightgrey","brown4","yellow4","orange4"),
line.type=1, leg=TRUE, pos=c(range[length(range)-2], range[length(range)]), do.grid=TRUE,
plot.precision=TRUE, trap.rule=TRUE)

performance.curves.plot(m, x.range=seq(from=0.1, to=1, by=0.1), 
y.range=c(0,1), curve.names=1:nrow(m), cex.val=0.6, f="", height=9, width=11,
col=c("black","red1","blue1","green1","darkgrey","brown1","yellow1","orange1",
"red4","blue4","green4","lightgrey","brown4","yellow4","orange4"), line.type=1, 
patch.type=1:16, leg=TRUE, pos=c(x.range[length(x.range)-2], y.range[2]), do.grid=TRUE,
x.label="Recall", y.label="Precision")

Arguments

y	a list of lists. Each component list is a list returned from precision.at.all.recall.levels that reports precision and recall results at different levels for different methods or tasks
range	numeric vector of the precision/recall values to be represented (def: values between 0 and 1 step 0.1)
curve.names	names of the compared methods to be reported in the legenda (def: numbers)
cex.val	magnification value for characters (def. 0.6)
f	file name. If is given, an encapsulated postscript file is created, otherwise the output is rendered on a window.
height	relative heigth of the graph (def. 9)
width	relative width of the graph (def. 11)
col	colors of the lines. 14 different colors are given as default, but any vector of color from colors() (package graphics) can be used. Colors are recycled if length(col) < length(y).
line.type	type of the line. Any valid vector of integer can be assigned (values between 1 and 6, see lty in par(), package graphics for details). Values are recycled if length(line.type) < length(y). Def.: 1 (solid lines).
leg	boolean: if TRUE (def.) a legend is depicted.
pos	coordinates of the position of the legend.
plot.precision	boolean: if TRUE (def.) precision/recall curves are plotted, othewise f-score/recall curves.
trap.rule	boolean: if TRUE (def.) the integral of the curves are computed.
m	a numeric matrix. Rows correspond to different methods and columns to precision or f-score at given recall values
x.range	vector of the recall values to be represented
y.range	vector with 2 elements: range of the precision/f-score to be represented
patch.type	numeric vector corresponding to the symbols to be plotted for different recall values (def. 1:16)
do.grid	a bollen value indicating whether a grid will be plotted or not (def: TRUE)
x.label	label of the abscissa (def: Recall)
y.label	label of the ordinate (def: Precision)

Details

The function precision.recall.curves.plot plots multiple precision/recall curves (or f-score/recall curves, if the argument plot.precision=FALSE) by using the output of the function precision.at.all.recall.levels, that compute several precison/recall pairs by moving the threshold from the lowest to the highest score achieved by each example.

The function performance.curves.plot plots precision of F-score/recall curves ofr a predefined set of recall levels. Thsi function can be useful to plot and to compare the average results between methods across multiple classes.

The curves can differ by color, type of line and for performance.curves.plot for each recall value a symbol can be also plotted. A legend can be automatically constructed.

Value

The functions output a graphic file either on a window or on an encapsulated postscript file. The function precision.recall.curves.plot, if the parameter trap.rule = TRUE (default), outputs a vector with the AUPRC (or the Area Under the F-score Curve if the parameter plot.precision=FALSE) for each curve.

Examples

# loading  an example matrix of scores an the correponding table of classes
data(T);
data(Scores);
res=list();
classes=c(1,2,7,8)
# computing precison recall values
for (j in classes) res=c(res, list(precision.at.all.recall.levels(Scores[,j], T[,j])));
# plotting precision/recall curves
precision.recall.curves.plot(res,curve.names=colnames(T)[classes], 
pos=c(0.7,1), plot.precision=TRUE);
# black and white version
precision.recall.curves.plot(res,curve.names=colnames(T)[classes], pos=c(0.7,1), 
plot.precision=TRUE, line.type=1:4, col=1);
# plotting f-score/recall curves
precision.recall.curves.plot(res,curve.names=colnames(T)[classes], pos=c(0.7,1), 
plot.precision=FALSE);
# black and white version
precision.recall.curves.plot(res,curve.names=colnames(T)[classes], pos=c(0.7,1), 
plot.precision=TRUE, line.type=1:4, col=1);

PerfMeas documentation built on Sept. 14, 2022, 9:06 a.m.