Weka Tutorial 11 Generating Non-stratified Folds (Data Preprocessing) Video

hello and welcome to the 11th way
Commission learning tutorial in this
tutorial we're going to cover a question
when I'm using a key fold cross
validation can I get each of the folds
from Wakeham well this is a very good
question because when you're doing a
careful cross validation by using any
kind of classification algorithms what
you see in Waker is that you are getting
the final results by averaging the
performances of the classification
algorithms across the K folds or
sometimes this is very usual that as a
researcher or data miner you need the
folds of these cross-validation sets so
how can you get each of the folds from a
k-fold cross-validation set up from wake
up so we're going to start the wake
Explorer and I'm going to open a file
that's I'm currently working on so this
is our data file and you can see that we
have seven thousand six hundred and
twenty-nine instances and thirty-nine
attributes so now if you go to the
classify tab and if you choose the naive
Bayes classification algorithm and we're
doing a 10-fold cross-validation then if
you put start then use you can see that
we are getting the results here and
these results are the average of all the
ten folds okay so what wicked does is it
automatically generates ten volts
because we have ten here as the number
of volts and for each fold it has the
position recall if measure correctly
classified instances are incorrectly
classified instances and so on and
afterwards after running for all the ten
folds it produces an average result and
it displays it on the classifier output
pane
but sometimes as I said before you might
need that Falls that Weka is generating
for this particular task zero to get the
false you can go to the pre process tab
you can use the filter you can choose
the filters and from the filters you go
to the unsupervised instance filters and
you can see there is an option called
remove folds so when you select the
remove folds you can go and you can see
that this filter takes the data set and
outputs and specified fold for
cross-validation so in the full tab if
this is one that means that you are
going to generate the first fall for
this data set the inverse selection is
false that means that you are currently
actually trying to get the first fold
sorry first of all from your data set if
that is true that means that you are
going to generate the folds from false
ranging from two to nine two to ten and
here the number of folds is specified as
ten we are not using any kind of seed so
that's set to default zero so we're
going to make it false because we're
trying to get the first fold for this
data set and we click OK so if you if
you choose apply you can see that the
number of instances are now 763 so
that's the data that's that in his test
number of instances for your first fold
and you can always save this fold by
clicking on save and choosing a name so
you can see that this has been saved as
one dot arf they have to remember that
if you choose the unsupervised instance
filter for rent a remove false option if
you choose this filter then what you're
going to do is you are not actually
stratifying the folds what's a
stratification
what does stratification mean well
stratification means that when you're
using a training and test test folds or
you're just splitting your data set into
some some number of folds at that time
this is not guaranteed
at all the classes will be will be
presented in a particular fold so if
your training instances if your training
set has for example all the ten classes
but your test instances the test set you
are producing for by using this filter
that doesn't have all the ten classes
represented then your classification
performance will be very poor so in
order to stratify defaults you can use
the supervised filter and go to the
instance and here you have an option
called stratified remote false it's the
same as the unsupervised remove old
filter but the only difference is that
when you are choosing this the
stratified removable filter at that time
you're guaranteeing that every fold will
have all every class in your datasets so
I hope that helps and see you in the
next tutorial
Thanks