oneClass: One-class classification in the absence of test data

Description Usage Arguments Details Value Examples

Calculation of accuracy for specified thresholds. evaluate is called to perform the calculations.

1
2
3

evaluateOcc(x, te.u, te.y, positive = NULL, th = NULL, allModels = FALSE,
  modParam = NULL, modRow = NULL, modRank = NULL, by = NULL,
  decreasing = TRUE, verbose = FALSE, ...)

`x`	an object of class `trainOcc`
`te.u`	the data to predict on.
`te.y`	a vector of observed positive/negative (presence/absence) values
`positive`	the positive label in te.y. if this is not given, the class with the lower frequency is assumed to be the positive class.
`th`	a vector of thresholds at which the model is evaluated
`allModels`	logical, default is `FALSE`. Set to `TRUE` if all models should be evaluated.
`modParam`	data frame with the parameters of the model to be evaluated
`modRow`	the row of the model in the `x$results` table.
`modRank`	the rank of the model after sorting by `by`.
`by`	character. must be a metric available in the `x$results` table by which to sort. If `NULL` the performance metric is taken from the `train` object. see also `sort.train`
`decreasing`	only when `modRank` is used. `TRUE` (default) to sort in decreasing order. Can be a vector if `by` is a vector.
`verbose`	`FALSE`, set to `TRUE` if you want to see the progress
`...`	arguments passed to `evaluate` from the `dismo` package.

By defalut, only the final model is evaluated when allModels is FALSE and non of the arguments modParam,modRow, or modRank given.

an object of class ModelEvaluation (see ModelEvaluation-class)) or ModelSelectionEvaluation. The latter is a list with two elements, the first containing the model sleetion table and the second a list with the evaluation results, each of whith a ModelEvaluation object. The rows in model selection table correspond to the evaluation list elements.

## Not run: 
# get training and test data
data(bananas)
seed <- 123456
tr.x <- bananas$tr[, -1]
tr.y <- bananas$tr[, 1]
set.seed (seed)
te.i <- sample ( ncell (bananas$y), 1000 )
te.x <- extract (bananas$x, te.i)
te.y <- extract (bananas$y, te.i)
# run trainOcc 
oc <- trainOcc(x=tr.x, y=puFactor(tr.y), 
               tuneGrid=expand.grid(sigma=c(0.1,1), ### not so large grid
                                    cNeg=2^seq(-5, 10, 3), 
                                    cMultiplier=2^seq(4, 15, 2)))
# evaluate the final model
ev <- evaluateOcc(oc, y=te.y, te.u=te.x)
# besides the thresholds used, this is identical to: 
te.pred <- predict(oc, te.x)
ev <- evaluate(p=te.pred[te.y==1], a=te.pred[te.y!=1])
# evaluate several models
# e.g. evaluate models with a true positive rate (tpr) higher than 0.8 and a 
# positive prediction probability (ppp) small than 0.4
modRows <- which(oc$results$tpr>=0.8 & oc$results$ppp<0.4)
ev <- evaluateOcc(oc, y=te.y, te.u=te.x, modRow=modRows)
# plot the pu-performance metric versus the maximum kappa 
evList <- print(ev)
plot(evList$puF, evList$mxK.K, xlab="puF", ylab="max. Kappa")

## End(Not run)