we are trying to connect weka with Postgresql. we changed the file
DatabaseUtils.props, but when we try to open a database from GUI, a
message like: "No suitable Driver" appears.
What we have in the DatabaseUtils.props file is:
Is there anything else that we have to set for weka read the
Vania and Baris
I am working on a 3-class classification problem with the k-NN
classifier (IBk). I tried two different methods:
A - using IBk directly as a 3-class k-NN classifier
B - using the MultiClassClassifier with IBk as the base classifier.
Method B gave me better results than Method A, but I don't understand
why there is a difference in performance.
Looking at the source code, it seems MultiClassClassifier picks the
class with the highest probability of the test instance belonging to
the class. But that's the same as using IBk by itself (method A) and
the probabilies of belonging to a class should be the same in both
methods. Is the difference in performance because k (the number of
neighbors) is being optimized to different values in the 2 methods, or
is there something more fundamental causing the difference here?
If someone can share some insight on this puzzle, I'd really
appreciate it. Thanks.
>>I'm missing the NeuralNetwork.java classifier. /
>//>/ Is this because it is deprecated / no longer supported?
/>/> regards, Chris
/>I think it can be found under a different name:
You are absolutely right. Thank you very much. :D
please does anyone know where in the iBK class or other associated classes
in weka, where i can find the formular that computes similarity between the
test and training sets.
I mean the formular which iBK uses to compute similarity before retrieving
the k nearest neighbours during classification.
>To: miryama Emma <miryama1(a)hotmail.com>
>Subject: Re: wish to join the list
>Date: Wed, 27 Oct 2004 10:17:36 +1300
>Fill out the form on this page, and click 'subscribe'.
>miryama Emma wrote:
>>Please i wish to join the wekalist. Can i know the necessary steps to
>>follow to do that.
>>It's fast, it's easy and it's free. Get MSN Messenger today!
It's fast, it's easy and it's free. Get MSN Messenger today!
Just installed the linux version of Weka 3-4-3. i.e. no windowsinstaller
used. I'm missing the NeuralNetwork.java classifier. I searched all over for
it, but it is not present anywhere, except in the older versions of Weka.
Is this because it is deprecated / no longer supported?
Dear Weka users,
I am trying to apply Naive Bayes Multinomial algorithm on Reuters text data
sets. I preprocessed the documents in a standard way, i.e., stopwords
removing and stemming, and I get the preprocessed data in arff file. Since
I am doing binary classification, I chose the 5 most popular categories as
the positive classes and generate 5 different data sets with positive and
However, the results of Naive Bayes Multinomial from these 5 data sets are
strange. It seems to assign all the instances to the negative classes
because the number of incorrectly classified instances are equal to the
number of positive instances in all 5 cases. Does anyone know why it gives
this result? Thanks in advance!
Greetings fellow Weka users,
I am quite new to the field of data mining. I am
having problem creating the training,
cross-validation, and test sets from the original
data set? In the book, "Data Mining: Practical
Machine Learning Tools and Techniques with Java
Implementations", there was mention of randomly
creating the 3 data sets
(training,cross-validation,test sets) mentioned
above as well as a stratified technique that
divides the classes from the original dataset to
the 3 data sets in equal proportions. Can you give
me a step-by-step procedure for carrying out such
I appreciate all comments and advice.
This mail sent through MU-Webmail: webmail.mahidol.ac.th
This has probably been answered before but I can't find it.
I get slight differences between BayesNet and Naive Bayes when run on the
same dataset. The net that is built is nothing more than the result variable
as a parent to all of the other variables. This is effectively the model
used in Naive Bayes (if I understand things correctly). The difference isn't
large -- about 0.3-0.5 percent. I still find it puzzling.
I have missing data elements. Does BayesNet sample for missing items while
NaiveBayes ignores them?
I have some troubles using the retrieveInstances method (from the class
InstanceQuery) since I updated my WEKA version (from a very older one to
the 3.4.3, same trouble with 3.4.2 and 3.4.1).
The method throws this exception :
Here is the query I use to retrieve my instances:
"Select login, count(login) from countlogins group by login"
login is a varchar(50) attribute.
The problem sounds to come from the "count" as it works with other
Does anybody have ever tackle this problem??