roc_curves: Receiver Operating Characteristic (ROC) Curves
In lcmix: Layered and chained mixture models

Description Usage Arguments Details Value Author(s) Examples

Functions to calculate and plot Receiver Operating Characteristic (ROC) curves for one or more sets of predictions, and for one or more objects of class mixmod or mdmixmod.

## Default S3 method:
rocinfo(x, labels, quasi=FALSE, ...)
## S3 method for class 'mixmod'
rocinfo(x, labels, quasi=FALSE, ...)
## S3 method for class 'rocinfo'
plot(x, legend="x", cex.legend=1, auc=TRUE, dca=TRUE,
    col=1, lty=1, lwd=1, ylab="true positive rate",
    xlab=ifelse(x$quasi, "all positive rate", "false positive rate"),
    grid=FALSE, gridres=0.1, gridcol="lightgray", gridlty="dashed", ...)
## Default S3 method:
rocauc(x, labels, quasi=FALSE, ...)
## S3 method for class 'mixmod'
rocauc(x, labels, quasi=FALSE, ...)

multiroc(x, labels, quasi=FALSE)
## S3 method for class 'multiroc'
plot(x, legend=names(x), cex.legend=1, auc=TRUE, dca=FALSE, bw=FALSE,
    col=(if(bw) rep(1, length(x)) else 1:length(x)),
    lty=(if(bw) 1:length(x) else rep(1, length(x))),
    lwd=rep(1, length(x)), ylab="true positive rate",
    xlab=ifelse(x[[1]]$quasi, "all positive rate", "false positive rate"),
    grid=FALSE, gridres=0.1, gridcol="lightgray", gridlty="dashed", ...)

`x`	for `rocinfo` and `rocauc`, a vector of probabilities, or an object of class `mixmod` or `mdmixmod`; for `multiroc`, a list of objects suitable to be passed as arguments to `rocinfo`; for plotting methods, an object of the appropriate class.
`labels`	a vector of logical values of the same length as the probabilities, in which a value of `TRUE` indicates a true positive; for `multiroc`, optionally a list of such vectors of the same length, with the same element names, as `x`.
`quasi`	logical; if `TRUE`, the x-axis of the ROC curve denotes the total positive rate rather than the false positive rate.
`legend`	character or `FALSE`; if character, a legend will be drawn with the name(s) in `legend`; if `FALSE`, no legend will be drawn.
`cex.legend`	magnification to be used for legend text.
`auc`	logical; if `TRUE`, AUC, or Area Under the Curve, will be shown in the legend. Larger AUC indicates better performance. No effect if `legend` is `FALSE`.
`dca`	logical; if `TRUE`, DCA, or Distance of Closest Approach of the curve to (0,1), will be shown in the legend. Smaller DCA indicates better performance. No effect if `legend` is `FALSE`.
`bw`	logical; if `TRUE`, a black-and-white version of the plot will be drawn. Convenience function for drawing multiple ROC curves on the same plot.
`col, lty, lwd, ylab, xlab`	as the equivalent arguments to `plot.default` or `lines.default`. For `plot.multiroc`, `col`, `lty`, and `lwd` should be vectors of the same length as `x`.
`grid`	logical; if `TRUE`, draw a grid on the plot.
`gridres, gridcol, gridlty`	resolution, color, and line type for the grid. No effect if `grid` is `FALSE`.
`...`	further arguments to `plot.default` or `lines.default` for the plotting methods, no effect for the other functions.

rocinfo calculates the ROC curve x coordinates (false positive rate, or all positive rate if quasi is TRUE), y coordinates (true positive rate), AUC, and DCA for the given probabilities and labels. multiroc performs the same function for multiple probabilities, and optionally multiple labels.

For rocinfo, a list of class rocinfo having the following elements:

`tpr`	the true positive rate in ascending order of the probabilities.
`fpr`	the false positive rate in ascending order of the probabilities.
`apr`	the all positive rate in ascending order of the probabilities.
`auc`	the Area Under the Curve.
`dca`	the Distance of Closest Approach.
`quasi`	logical; the same as the argument to `rocinfo`.

For rocauc, the numerical value of the AUC.

For multiroc, a list of class multiroc having elements of class rocinfo.

The plot and lines methods are used for their side effects.

Daniel Dvorkin

## Not run: 

### multiple ROC curves for a single set of labels

data(CiData)
data(CiGene)
CiJoint <- lapply(namedList("layered", "chained"), function(top)
    mdmixmod(CiData, c(2,3,2), family=c("pvii", "norm", "pvii"),
    topology=top))
CiMarginal <- marginals(CiJoint$layered) # same for $layered and $chained
CiFits <- c(CiJoint, CiMarginal)
sapply(CiFits, rocauc, labels=CiGene$target)
#     layered      chained      binding   expression conservation 
#   0.8507258    0.8913765    0.8596395    0.7943286    0.7479479
plot(multiroc(CiFits, CiGene$target), grid=TRUE)

### multiple ROC curves for multiple labels

DlFits <- mdmixmod(DlData, c(2,2,2), family=c("norm", "norm", "pvii"))
AllFits = list(Ci=CiJoint$chained, Dl=DlFits)
plot(multiroc(AllFits, list(Ci=CiGene$target, Dl=DlGene$target)))

## End(Not run)