Weka Tutorial 07: Models 101 (Model Evaluation) Video Lecture | Weka Tutorial - Data & Analytics

to everyone now in this you will wake a
tutorial I'm going to show you something
that I myself didn't know before arm I
came across in one of the posts in Stack
Overflow someone over there posted a
question like this when you are using a
training set you were you can develop a
model and you can save that model and
when you're using a 10-fold
cross-validation you can also develop a
model and you can save that model so the
question is are those models same what
in conventional machine learning problem
the community that reports work on
10-fold cross-validation or
experimentation on training sets they
all the time report the model of
Ortonville cross-validation that
performs the based or sometimes they
also report that in case of 10-fold
cross-validation actually ten models are
developed and then finally you have the
average of those ten models as your
final model but for wake up that's no
more the case so for for in case of wake
up what it does for 10-fold
cross-validation is that it develops ten
models for ten foals but finally when it
displays the results it uses the average
performances of those ten models and
deletes all those models this is very
unfortunate for us because you do not
have the model on 10-fold
cross-validation anymore so what you are
having here at this stage the model you
have is actually developed from the
training set by using the entire
training set so I'm giving a giving you
an example here so in this tutorial I'm
going to use our way the dot Marfa
for our experimentation so we loaded
that and we go to the classify tab we
are going to use nine base and test
options we select use training set we
click on start so now we have a model
here at this result list pane and we are
going to save that model somewhere in
our hard disk so I'm going to save that
as training dot model now I'm changing
my test options to cross-validation with
ten volts and again I click start with
naive Bayes classifier you can see that
these two models have different of
measures precision and rickles so this
one was develop from 10-fold
cross-validation we have developed this
model and we are going to save that
model as CV don't model so in case of
10-fold cross-validation what Wei Chi is
doing here is it uses all the ten foals
it develops ten different models for ten
different folds and produces results and
displays them in the classifier output
PEM of wake up but after that waker did
its all ten models for ten foals and
what you are having here at this results
list pane this model is created at 7:13
p.m. this is the exact same model as
this one created as 712 p.m. so in order
to validate that statement you can see
the properties of our training model and
this is 3.9 kilobytes in size and you
can also check the properties of our CV
dot model and you can see that it has
the same exact same size three point
nine zero kilobyte so to summarize I can
say that in case of Waker when you're
using 10-fold cross-validation what
you're having here and this pane
it's the average performance results for
all the ten folds but when you are
saving this model the model waker has
created here is the exact same model if
you use the entire training set in order
to develop that model so you don't have
any options for saving the best model
either the best model or the average
model if you're using 10-fold
cross-validation thank you