Jose Lopez Prato wrote:
Hi... I am starting using weka. Here two questions.
The weka.classifiers.meta.Cleanser class has a -C option to use any
in the cleaning process. I understood, from the book you know, that the
algorithm of detect/remove erroneous
instances only works if we use decision trees because of the pruning
tree procedure. Obviously, I´m wrong..Why?
The technique has been shown to work well with decision trees, but it
can be used with any classifier that misclassifies some of the training
I am running the Cleanser class using the
The idea is to get from that running a clean training file.
So, if after the cleaninig I make a test over the same model that fixed
errors but using the new training file
I´d hope a fully right classified instances set. Why I cannot get that?.
Are you asking for the finally cleansed data set produced by the
procedure or asking why the set is still misclassified? If you are
asking for the set, at the moment it is used internally by the Cleanser
classifier but cannot be output. It should be possible to write a filter
that follows the cleansing procedure and returns the cleansed data set.