I am using Weka in Ubuntu11.10.
But don't know how to enlarge Weka GUI font size.
(It is in agony to read tiny font.)
I have search the Mail list archive, but can't find a solution.
Any answer will be appreciated!
I have a classifier that is generalizing poorly from its training
set to a test set, and I'm trying to debug.
The ability to list predictions for each instance is useful.
but when i have something like a j48 decision tree, i can imagine
these annotating the tree as the original training instances do:
flagging each leaf with the number of true/error predictions
(and even contrasting these with error rates from the original
has anyone tried to do something like this? any suggestions
as to where in the code i might graft something like this?
so far i've not even found the bit that generates the
ASCII-art decision tree:
> att1 = 0
> | att2 = 0
> | | att3 = 0
> | | | att4 = 0
> | | | | | | ... att5 = 0: 1 (382.0/153.0)
thanks for any pointers,
Due to out of memory error, i change the heap size in Runweka.ini from 1024M
to 2048M, after that i tried to start weka, but i could not. The i reduce
the heap size to 1400M and weka gets started.
How we can increase the heap size to enable weka to handle larger datasets.
View this message in context: http://weka.8497.n7.nabble.com/Weka-3-7-Heap-Size-tp36191.html
Sent from the WEKA mailing list archive at Nabble.com.
I (stupidly, in retrospect) installed an external package (NeuralNetwork)
via the package manager which broke my Weka installation. I now get the
menu for GUIChooser but nothing further happens.
I'm trying to uninstall NeuralNetwork via the command line, but it fails.
java weka.core.WekaPackageManager -list-packages installed
1.0.1 1.0.1 Yes naiveBayesTree: Class for generating a decision
tree with naive Bayes classifiers at the leaves.
1.0 ----- Yes NeuralNetwork: Neural Network
1.0.3 1.0.3 Yes NNge: Nearest-neighbor-like algorithm using
non-nested generalized exemplars (which are hyperrectangles that can be
viewed as if-then rules)
java weka.core.WekaPackageManager -uninstall-package NeuralNetwork
I'm running weka-3-7-13 (with the Apple JVM) on El Capitan.
Any thoughts as to how I can fix this?
On a side note, attempts to download a fresh copy via
fail. The link loops back to itself.
Thanks in advance for any assistance anyone can offer.
the current behavior of the InputMappedClassifier.toString() is to
pass off the work to the m_Classifier's own toString() method;
in the case of J48, this resolves to J48.ClassifierTree.dumpTree().
but dumpTree() contains hard-wired references to the TRAINING
Instances of m_train.
i'm wondering how you would recommend tweaking this code
to allow dumpTree() to produce the same output with respect
to the TESTING instances maintained in
i can only think of pretty gross work-arounds that require
some explicit dumpWrappedJ48tree() or similar.
thanks once again for any hints,
I am wondering if exists a non-negative least squares algorithm that learns
regression coefficients that only can be Natural numbers?
I would like to use it because I know that all my regression coefficients
are Natural numbers and therefore, perhaps I could obtain better results
than learning Rational numbers.
Thanks in advance,
View this message in context: http://weka.8497.n7.nabble.com/Non-negative-AND-NATURAL-least-squares-algor…
Sent from the WEKA mailing list archive at Nabble.com.
I'm working with a dataset where 3/5th of the variables are comprised of
binary values, 1/5 are comprised of numeric values and the remaining 1/5th,
Right now, I need to do me some good feature selection.
I've traditionally used infogainattributeeval
which is supposed to be able to discretize numerics and manage these
different data types on its own.
However, this time around, I tried out multiple other feature selection
methods,and found that others, such as CfsSubsetEval
perform significantly better.
Is there some rules on which feature selection method should be used on a
combination of discrete / numeric variables? or is it more 'try out
multiples, and use whatever performs best with your dataset?'
I've tried to figure out the best / most logical approach, but haven't
reached any proper conclusions...
I general I need a better understanding of how to interpret the results of
the Multilayer Perceptron. In particular I need to convert the results from
the Multilayer Perceptron to java code. It seems that i should use the
linear node output values but without the sigmoid node values I don't see
how that can work. If i use the sigmoid nodes then I guess I would use the
combination of values and their positive or negative values to determine
the class results for that range. Any assistance you can provide would be
I am a PhD researcher working in the area of requirements engineering. I am trying to use WEKA to cluster the weights assigned by stakeholders' on some lists of requirements specification for a Pharmacy Information System. The lists of requirements were elicited from the stakeholders (who are pharmacist), and one goal of the research is to resolve conflicting expectations of these weights assigned by each stakeholders'.
The major challenge I have is how to interpret the results, and also the visualization of the clusters. Attached is the dataset I used in CSV format.
In the dataset, the R1 to R101 are the requirements ID representing the lists of requirements. The numbers 5, 4, 3, 2, 1 are the weights assigned by each stakeholder on each requirement. In all, the dataset contains 42 responses (i.e. 42 stakeholders assigned weights to each requirement) and 101 requirements. On clustering the dataset using 5 clusters, it displayed 5 clusters with their centroids/mean and percentages assigned to each cluster. How do I interpret this result? What should I consider in visualizing these clusters? Please, I need assistance.