function to build objects of class 'rocc'
rocdemo.sca – demonstrate 'rocc' class construction using a scalar marker and simple functional rule
rocdemo.sca(truth, data, rule=NULL,
cutpts=NA,
markerLabel="unnamed marker", caseLabel="unnamed diagnosis",
quiet=TRUE)truth |
true classification of objects. Must take values 0 or 1. |
data |
quantitative markers used to classify |
rule |
rule: a function with arguments (x, thresh) returning 0
or 1. If no rule is provided or the standard rule |
cutpts |
values of thresholds; no NA allowed, or they will be recomputed using smallest gap between data points with distinct values |
markerLabel |
textual label describing marker |
caseLabel |
textual label describing classification |
quiet |
defaults to TRUE, suppressing message about discovery of NA in cutpts |
dxrule.sca is function (x, thresh) ifelse(x > thresh, 1, 0)
The default value of argument cutpts is a point less than min(data), points separating the unique values of data and a point greater than max(data).
an object of S4 class rocc
Vince Carey (stvjc@channing.harvard.edu)
AUC
set.seed(123)
R1 <- rocdemo.sca( rbinom(40,1,.3), rnorm(40), caseLabel="new case", markerLabel="demo Marker" )
plot(R1, line=TRUE, show.thresh=TRUE)
truth <- c(0, 1, 0, 1, 1, 0, 1, 1)
data <- c(2, 3, 4, 4, 5, 6, 7, 8)
R2 <- rocdemo.sca(truth, data, dxrule.sca)
plot(R2, line=TRUE, show.thresh=TRUE)
R3 <- rocdemo.sca(truth, data, function(x, thresh) 1 - dxrule.sca(x, thresh))
if (AUC(R2) + AUC(R3) != 1) stop('Sum of AUCs should be 1.')
#
# more involved
#
set.seed(1234)
x = runif(1000)
w = runif(1000)
z = rbinom(1000, 1, plogis(-2.7+6.2*x + .3*w))
m1 = glm(z~x, fam=binomial)
demorule.glm.clo = function(model) function(w,thresh)
ifelse(predict(model, newdata=list(x=w), type="response")>thresh, 1, 0)
demorule.glm = demorule.glm.clo(m1)
R4 = rocdemo.sca(z, x, demorule.glm )
plot(R4)Please choose more modern alternatives, such as Google Chrome or Mozilla Firefox.