hldEstimates: Performance estimation using holdout and random resampling
In performanceEstimation: An Infra-Structure for Performance Estimation of Predictive Models

Description Usage Arguments Details Value Author(s) References See Also Examples

This function obtains hold-out and random sub-sampling estimates of performance metrics for a given predictive task and method to solve it (i.e. a workflow). The function is general in the sense that the workflow function that the user provides as the solution to the task, can implement or call whatever modeling technique the user wants.

The function implements hold-out and random sub-sampling (repeated hold-out) estimation. Different settings concerning this methodology are available through the argument estTask (check the help page of Holdout).

Please note that most of the times you will not call this function directly (though there is nothing wrong in doing it) but instead you will use the function performanceEstimation, that allows you to carry out performance estimation for multiple workflows on multiple tasks, using some estimation method you want (e.g. hold-out). Still, when you simply want to have the hold-out estimate of one workflow on one task, you may prefer to use this function directly.

1	hldEstimates(wf,task,estTask,cluster)

`wf`	an object of the class `Workflow` representing the modeling approach to be evaluated on a certain task.
`task`	an object of the class `PredTask` representing the prediction task to be used in the evaluation.
`estTask`	an object of the class `EstimationTask` representing the hold-out settings to use.
`cluster`	an optional parameter that can either be `TRUE` or a `cluster`. In case of `TRUE` the function will run in parallel and will internally setup the parallel back-end (defaulting to using half of the cores in your local machine). You may also setup outside your parallel back-end (c.f. `makeCluster`) and then pass the resulting `cluster` object to this function using this parameter. In case no value is provided for this parameter the function will run sequentially.

The idea of this function is to carry out a hold-out experiment with the goal of obtaining reliable estimates of the predictive performance of a certain approach to a predictive task. This approach (denoted here as a workflow) will be evaluated on the given predictive task using some user-selected metrics, and this function will provide hold-out or random sub-sampling estimates of the true value of these evaluation metrics. Hold-out estimates are obtained by randomly partition the given data set into train and test sets. The training set is used to obtain a model for the predictive task, which is then tested by making predictions for the test set. This random split of the given data can be repeated several times leading to what is usually known as random sub-sampling estimates. In the end the average of the scores over the several repetitions (if using pure hold-out this is only one) are the hold-out estimates of the selected metrics.

Parallel execution of the estimation experiment is only recommended for minimally large data sets otherwise you may actually increase the computation time due to communication costs between the processes.

The result of the function is an object of class EstimationResults.

Luis Torgo ltorgo@dcc.fc.up.pt

Torgo, L. (2014) An Infra-Structure for Performance Estimation and Experimental Comparison of Predictive Models in R. arXiv:1412.0436 [cs.MS] http://arxiv.org/abs/1412.0436

Holdout, Workflow, standardWF, PredTask, EstimationTask, performanceEstimation, cvEstimates, bootEstimates, loocvEstimates, mcEstimates, EstimationResults

## Not run: 

## Estimating the mean absolute error and the normalized mean squared
## error of rpart on the swiss data, using 70%-30% hold-out
library(e1071)
data(swiss)

## Now the evaluation
eval.res <- hldEstimates(
             Workflow(wf="standardWF",wfID="svmApproach",
                      learner="svm",learner.pars=list(cost=10,gamma=0.1)
                     ),
             PredTask(Infant.Mortality ~ ., swiss),
             EstimationTask(metrics=c("mae","nmse"),
                            method=Holdout(nReps=5,hldSz=0.3))
                        )

## Check a summary of the results
summary(eval.res)

## An example with a user-defined workflow function implementing a
## simple approach using linear regression models but also containing
## some data-preprocessing and well as results post-processing.
myLM <- function(form,train,test,k=10,.outModel=FALSE) {
    require(DMwR)
    ## fill-in NAs on both the train and test sets
    ntr <- knnImputation(train,k)
    nts <- knnImputation(test,k,distData=train)
    ## obtain a linear regression model and simplify it
    md <- lm(form,ntr)
    md <- step(md)
    ## get the model predictions
    p <- predict(md,nts)
    ## post-process the predictions (this is an example assuming the target
    ## variable is always positive so we change negative predictions into 0)
    p <- ifelse(p < 0,0,p)
    ## now get the final return object
    res <- list(trues=responseValues(form,nts), preds=p)
    if (.outModel) res <- c(res,list(model=m))
    res
}

## Now for the Holdout estimation 
data(algae,package="DMwR")
eval.res2 <- hldEstimates(
             Workflow(wf="myLM",k=5),
             PredTask(a1 ~ ., algae[,1:12],"alga1"),
             EstimationTask("mse",method=Holdout(nReps=5)))

## Check a summary of the results
summary(eval.res2)

## End(Not run)