[R] cforest and cforest_unbiased for testing and training datasets

dub2 dub2 at illinois.edu
Thu Sep 13 21:29:32 CEST 2012


Greetings,
I am using cforest to predict age of fishes using several variables; as it
is rather difficult to age fishes I would like to show that a small subset
of fish (training dataset) can be aged, then using RF analysis, age can
accurately be predicted to the remaining individuals not in the subsample.
In cforest_unbiased the samples are drawn without replacement and so it
creates a default testing dataset (approx 35%) and training dataset from the
rest. My question is that if I have already separated my data into a testing
and training dataset prior to RF analysis is there a reason I should not set
the fraction option to .99 to essentially eliminate the testing dataset and
ensure I have the desired # of samples in my testing dataset? The resulting
analysis will be applied to my specified testing dataset to assess
misclassification and accuracy. I am aware that the options of
cforest_unbiased do not allow allow the fraction value to be altered so my
second question is is data.controls <- cforest_control(teststat = "quad",
testtype = "Univ", replace = FALSE, ntree=2000, mtry=2, fraction=.99) the
equivalent to data.controls <- cforest_unbiased(ntree=2000, mtry=2) except I
have eliminated the default testing dataset?
Thank you for your help 



--
View this message in context: http://r.789695.n4.nabble.com/cforest-and-cforest-unbiased-for-testing-and-training-datasets-tp4643057.html
Sent from the R help mailing list archive at Nabble.com.



More information about the R-help mailing list