You may be able to do this in the KnowledgeFlow somehow, not sure, but I
would probably write a Groovy or Jython script and run it in the
corresponding console (available from the Tools menu in the GUIChooser if
you have the kfGroovy and tigerJython packages installed).
> ---------- Forwarded message ----------
> From: Rana Sms <rana.communication.ahmed(a)gmail.com>
> To: "Weka machine learning workbench list." <wekalist(a)list.waikato.ac.nz>
> Cc: rana(a)eri.sci.eg
> Date: Tue, 30 Jan 2018 18:54:46 +0200
> Subject: Read folder of arff files
> I have a folder of 100 arff file. And already built random forest model
> How could i evaluate the performance of all these files automatically and
> make weka reads the arff folder serially without need to test each file
How to pre-process the data in order to have zero mean and unit variance and to discretize the data into categories to reduce noise.
I tried the "MultiFilter", using (first filter) the Unsupervised Standardize (to have zero mean and unit variance) and then I applied the MathExpression (second filter) in order to discretize the data into categories using the IfElse syntax, but it didnt work with me. At the end I am getting the discritization result but the mean is not equal to zero (standardize is not working).
Can any one help.
I have an issue regarding selecting attributes. I am doing a study to compare selection methods. When I apply the "MultiFilter" in the Pre-process panel using supervised discretization and genetic search, I am getting a different result when I use the Classify Panel and applying the "FilteredClassifier" using the same multifilter methods (discrtize+ genetic search).
I don't mean here about the classification accuracy, I am surprised by the different number of attributes. But when I used other methods, I am getting the same number of attributes. Can anyone help to clarify this for me.
I have an unlabeled 20 newsgroup text dataset without class labels. But I
don't know class labels; I want to use weka API to classify it. But weka
uses class labels to classify datasets. The link is 20ng dataset
<http://csmining.org/index.php/id-20-newsgroups.html>. I am quite new to
datamining, so don't know much about assigning class labels etc. I need to
classify this dataset and want to convert it into arff format but don't
have idea; I know arff format but need an idea so as, how to convert this
dataset so that weka could accept it and I can classify it.
No, there is currently no way to change the precision in the output. You
would need to write some Java code or similar to access the values in the
From: hihichew <hihichew(a)gmail.com>
> To: wekalist(a)list.waikato.ac.nz
> Date: Fri, 26 Jan 2018 00:42:42 -0700 (MST)
> Subject: Decimal Points in Results <Error on test data / Detailed Accuracy
> By Class>
> is there any way to toggle the decimals in both "Error on test data /
> Detailed accuracy by class" via the CLI?
> i have looked into the command of num-decimal-places (As far as i know,
> command just changes the number of decimal places when a model is being
> built.) Besides, i have also tried the 'output prediction' decimal from the
> weka explorer. (nothing seems to be working)
> The main reason i am trying to do this is because:
> The weighted average of my FP rate is equivalent to 0, but through manual
> calculation based on the confusion matrix
> (i am able to get some of the value for around 0.0004, 0.0005.... )
> my current command:
> java -Xmx20000m -cp weka.jar weka.classifiers.trees.J48 -l
> -T "25_Ori_Gure.arff" > results.txt
> Sent from: http://weka.8497.n7.nabble.com/
I had to perform regression for some set of data with Weka Java API. The
objective of regression was attribute with numeric class. Before regression,
I had performed select attributes. From this set of attributes, I extracted
subsets with 2,3,5,15 and 25 best of 28 attributes and set this attributes.
>From training and test data I filtered out other attributes, and left only
best ones and class attribute. On training data I had build classifiers
(MultilayerPerceptron, M5P, LinearRegression, RandomForest, Bagging of
MultilayerPerceptrons and Bagging of M5P). The classifier was built on
prepared set of training data (with filtered out attributes, which were not
selected in attributes selection), and then at all record of test data i run
classifyInstance. When I got the predictions, the best results were
evaluated for subset with 15 attributes.
My question is: *if classifier objects has some internal feature selection?*
I had not changed anything when creating classifiers - the default
constructors were used.
Thanks in advance,
Sent from: http://weka.8497.n7.nabble.com/
I would like to compute the logarithm of base 10 in weka. I noticed that log function in the MathExpression function computes the Natural Logarithm (ln). I tried to used the "log10(A)" function as in Java but has an error. Can anyone help.