cv52 {gcl}R Documentation

5x2 fold crossvalidation

Description

Function to perform 5x2-fold crossvalidation comparison of two predictive models on the same data.

Usage

cv52(df, mb1, mb2, eval, seed = NULL, cv.verbose = F, ...)

Arguments

df The data frame, last column is outcome
mb1 Model builder 1, must return a function that takes df as input and returns a matrix where entry (i,j) is the classifiers belief of case i's memebership in class j
mb2 Model builder 2, must return a function that takes df as input and returns a matrix where entry (i,j) is the classifiers belief of case i's memebership in class j
eval Evaluation function, takes as input classifier generated by mb[12] above and data frame df
seed Random seed.
cv.verbose If TRUE, spew info.
... Passed on.

Details

This function implements the 5x2 CV F statistic proposed by Alpaydin as an improvement of Dietterich's statistic.

Value

Returns a list.

f the F statistic value
p p value
m matrix of split/fold results
seeds vector of random seeds used

Author(s)

Staal A. Vinterbo staal@dsg.harvard.edu

References

Alpaydin: "Combined 5x2 cv F test for comparing supervised classification learning algorithms" Neural Computation 1999

See Also

cv,acc.eval,ci.eval

Examples

## Not run: 
cv52(df, gcl, tcl, acc.eval)
## End(Not run)

[Package gcl version 1.06.5 Index]