performanceEstimation (version 1.1.0)

cvEstimates: Performance estimation using cross validation

Description

This function obtains cross validation 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 N x K-fold cross validation (CV) estimation. Different settings concerning this methodology are available through the argument estTask (check the help page of EstimationTask and CV).

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 the estimation method you want (e.g. cross validation). Still, when you simply want to have the CV estimate of one workflow on one task, you may prefer to use this function directly.

Usage

cvEstimates(wf,task,estTask,cluster)

Arguments

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 defining the prediction task for which we want estimates.

estTask

an object of the class EstimationTask indicating the metrics to be estimated and the cross validation 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.

Value

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

Details

The idea of this function is to carry out a cross validation 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 k-fold cross validation estimates of the true values of these evaluation metrics. k-Fold cross validation estimates are obtained by randomly partitioning the given data set into k equal size sub-sets. Then a learn+test process is repeated k times. At each iteration one of the k partitions is left aside as test set and the model is obtained with a training set formed by the remaining k-1 partitions. The process is repeated leaving each time one of the k partitions aside as test set. In the end the average of the k scores obtained on each iteration is the cross validation estimate.

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.

References

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

See Also

CV, Workflow, standardWF, PredTask, EstimationTask, performanceEstimation, hldEstimates, bootEstimates, loocvEstimates, mcEstimates, EstimationResults

Examples

## Not run: 
# 
# ## Estimating the mean squared error  of svm on the swiss data,
# ## using two repetitions of 10-fold CV
# library(e1071)
# data(swiss)
# 
# ## Now the evaluation
# eval.res <- cvEstimates(
#              Workflow(wf="standardWF", wfID="mySVMtrial",
#                       learner="svm", learner.pars=list(cost=10,gamma=0.1)
#                      ),
#              PredTask(Infant.Mortality ~ ., swiss),
#              EstimationTask(metrics="mse",method=CV(nReps=2,nFolds=10))
#                        )
# 
# ## 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 CV estimation 
# data(algae,package="DMwR")
# eval.res2 <- cvEstimates(
#              Workflow(wf="myLM",k=5),
#              PredTask(a1 ~ ., algae[,1:12],"alga1"),
#              EstimationTask("mse",method=CV()))
# 
# ## Check a summary of the results
# summary(eval.res2)
# 
# ##
# ## Parallel execution example
# ##
# ## Comparing the time of sequential and parallel execution
# ## using half of the cores of the local machine
# ##
# data(Satellite,package="mlbench")
# library(e1071)
# system.time({p <- cvEstimates(Workflow(learner="svm"),
#                               PredTask(classes ~ .,Satellite),
#                               EstimationTask("err",Boot(nReps=10)),
#                               cluster=TRUE)})[3]
# system.time({np <- cvEstimates(Workflow(learner="svm"),
#                                PredTask(classes ~ .,Satellite),
#                                EstimationTask("err",Boot(nReps=10)))})[3]
# 
# ## End(Not run)