cv52 {gcl} | R Documentation |
Function to perform 5x2-fold crossvalidation comparison of two predictive models on the same data.
cv52(df, mb1, mb2, eval, seed = NULL, cv.verbose = F, ...)
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. |
This function implements the 5x2 CV F statistic proposed by Alpaydin as an improvement of Dietterich's statistic.
Returns a list.
f |
the F statistic value |
p |
p value |
m |
matrix of split/fold results |
seeds |
vector of random seeds used |
Staal A. Vinterbo staal@dsg.harvard.edu
Alpaydin: "Combined 5x2 cv F test for comparing supervised classification learning algorithms" Neural Computation 1999
## Not run: cv52(df, gcl, tcl, acc.eval) ## End(Not run)