Surendra Singhi wrote:
10fold cross validation means that a test is done
10 times. Everytime
with another random 10% of the data.
Who said it is random? Cross-validation is not done with random 10% of
In fact the data is first stratified and then divided into 10(num_folds)
folds and then out of these folds in each turn separate 9
folds(num_folds-1) are used for training and the remaining 1 fold for
The results of the 10 tests are given as an
output by WEKA> these are
But averages aren't very usefull without any information on how the 10
seperate tests performed... Are all 10 tests almost the same? Or does
one score 100% and the other 0%?
Hope the above has answered all your questions.
>> Kazz wrote:
>> Actually the book of Frank and Witten says the data is divided in 10
>> parts using a random-function. Then 9 parts are used for training
>1 part for testing. This is done 10 times, everytime with another part
>as the test-part.
>But the divion in 10 parts is actually random.
>The problem I had is already solved though. The
Experimenter gives all
>percentages of the seperate 10 tests which makes it possible to
>calculate the standard deviation!
I might not have been clear, in my earlier message, actually the data is
first randomized, then stratified and then divided into 10 folds.
Hope this helps.
And can anyone please tell me why are people emailing me instead of
replying on the newsgrou?