I would like to know whether it is possible to manually construct (or)
edit a classifier model (say a Naive Bayes model by directly specifying
the prior probabilities).
If there are no APIs, has anyone come across any open source tool (that
uses Weka) with similar functionality?
Thanks in advance!
I want to multiply the information gain by a weight factor.
I made XRFF file by saving ARFF file as XRFF and then edit it with notepad and wrote in it after each attribute for example
<attribute name="sex" type="nominal"> <metadata> <property name="weight">0.9</property> </metadata> <labels> <label>female</label> <label>male</label> </labels> </attribute>
Is this XRFF file correct or not?
Then I changed in the code of both Id3 and J48 .
I changed in C45ModelSelection
Attribute weight = new Attribute ("weight"); //object from class Attribute
averageInfoGain = (averageInfoGain + currentModel[i].infoGain()) * weight.weight(); // ( 2 lines)
min Result = currentModel[i].gainRatio() * weight.weight();
When executing and running no change in the result appeared.
Is this correct or not?
return infoGain * att.weight();
When executing and running change appeared in the result.
Is this correct or not?
For J48 when running the classifier it gave number of leaves and the size of the tree =1 but when using the filter of instance resampling it appeared a tree with number of leaves = 79 and size of the tree = 116.
Why It cannot give the tree to me before resampling?
I have an imbalanced dataset.
I'm having issues figuring out how best to deal with this.
Does anyone have a way to deal with this in Weka that they would recommend?
*School of Computer Science and Statistics*
*Trinity College Dublin*
*e: leavys(a)tcd.ie <geneSIS(a)tcd.ie>***
I'm trying to build a multi-threaded application using Weka as a library.
Unfortunately, Weka does not call Thread.sleep() or check
Thread.isInterrupted() when doing long-running tasks such as
That means I cannot cancel or interrupt my own threaded jobs if they run
too long (which is a requirement).
Does anyone have any advice? Seems to me like an issue that someone might
have solved it before. Is there any other solution besides patching Weka
myself or calling out to it as a standalone application?
I have been struggling with this some more and am trying to adjust my
I want to apply n many learning algorithms to my training set using each of
the columns of ECOC and be able to use these individually. So if the ECOC
table is 15 wide and I use 4 learning algorithms I would have 60 total
I then want to then test these classifiers by reusing the training set as
test set. This will give me an output that should ideally resemble what
ECOC would expect but may have differences from time to time. The
classifier will spit out a binary value for each of the instance.
I'll then use that output to try to figure out which learning algorithm to
use for which column by trying to maximize column separation.
Basically in the past I tried letting WEKA pick the "best" classifier for
the situation, now I want to come up with a classifier for that purpose.
-- とある白い猫 (To Aru Shiroi Neko)