I'm using Weka ThresholdCurve Evaluation class to plot ROC curve, for my
Actually I'm using wekas threshold curve class to plot ROC curve from MOA
I'm dealing with a problem where in only 2 classes are present.
Now when this class *weka.classifiers.evaluation.ThesholdCurve* plots the
How do we know that which class is considered as Positive and Which class
is considered as Negative..?
while plotting the ROC..??
---- Abhijeet Godase.
I didn't understand how the resample is working in weka?when using it more sense?
I've got 400 instances related to 4 classes first 100 to class 1 second 100 to class 2 .....etc
i used it and it improved my classification results .but I wonder if using it according to my data logically accepted or not since I did not randomize my data ????? plzzzzzzzzz help
First post to this list. My question(s) will be somewhat general. I have
little background in machine learning.
My research is in the area of computer graphics. We are currently
considering how to apply Bayesian Networks to some of what we're doing. A
couple of papers that have given some inspiration are as follows:
Now, consider a basic problem where we have a set of input polygons (2D). I
want to classify these shapes based on some descriptors. My question is
whether using Weka (admittedly as a black-box approach given my lack of
background) will be helpful. I will try to explain the process and would be
grateful for any feedback or comments:
1. Generate a collection of 2D rectangular shapes.
2. Create a vector-representation of each shape.
3. Use Weka's attribute-selection algorithm(s) to determine the most
4. Use Weka to generate a Bayesian Network (since we will have a small set
of input shapes). The hope is that we can generate new unseen shapes that
belong to a desired class.
This email could get long... so I'm not sure whether I should stop here or
continue. But I do want to give an example to clarify:
Consider the three shapes that are attached to this email. Those shapes are
simply created by offsetting a set of rectangles from the origin, and then
tracing the boundary of the set of those rectangles (in each case). So a
simple vector representation for each shape (1, 2, and 3) could be:
1) 75.0 50.0 0.0 0.0, 25.0 25.0 20.0 20.0, 35.0 25.0 -20.0 -20.0,
2) 50.0 75.0 0.0 0.0, 25.0 25.0 20.0 20.0, 35.0 25.0 -20.0 -20.0,
3) 75.0 40.0 0.0 0.0, 60.0 20.0 -7.0 20.0, 27.0 55.0 24.0 -10.0,
22.0 25.0 -26.0 -20.0,
(Each rectangle is represented by [width, height, offset-x, offset-y].)
Assume that there is a larger set of such shapes (e.g., 100). Would this be
something that I could feed to Weka, and would it be able to determine if
there are any parameters that describe the class of these shapes? (Suppose
that these three shapes belong to the class "feasible" and then I could
generate some undesirable shapes that belong to the class "unfeasible".)
>From there I would hope to be able to construct a Bayesian Network and
eventually generate new shapes belonging to the "feasible" class.
I realize this may sound like a silly example, but I'm really trying to
understand process here and would be grateful for any feedback. I've
already done a tutorial on Weka that somewhat follows this pattern using
By default, for a regression problem Weka outputs the following values:
Correlation coefficient, Mean absolute error, Root mean squared error,
Relative absolute error, Root relative squared error, Total Number of
1. Where (in what class) could I see how they are computed?
2. How could I add my own measurement function to be outputted in the Weka
GUI, along with the ones above?
View this message in context: http://old.nabble.com/Adding-a-new-evaluation-metric-tp33236997p33236997.ht…
Sent from the WEKA mailing list archive at Nabble.com.
When I use libsvmsaver via the weka gui it works just fine. But when I use it via the command line with the -c first argument it does not work. I tried all kinds of things. -c 1 -c "first". It still takes the last row as the class attribute.
Aram Hovsepyan, Ph. D.
Department of Computer Science
B-3001 Leuven (Belgium)
Phone: +32 (0) 16/32.79.55
I just want to make sure that I understood the algorithm of SOM correctly since I need to modify the code a bit.
A clustering result of SOM is based on the result of last epoch, right?
Passing through different epochs, the weight vector of each neuron changes and I suppose the BMU for each input also changes.
However a clustering result only refers to the final round of epoch after all changes right?
Then I am just wondering if I use the clustering membership filter with SOM, how does it calculate the probability? which epoch it refers to?
Did I misunderstand something about the algorithm?
Am I clear about what I wanted to say?
Thank you in advance!
Este mensaje se dirige exclusivamente a su destinatario. Puede consultar nuestra pol?tica de env?o y recepci?n de correo electr?nico en el enlace situado m?s abajo.
This message is intended exclusively for its addressee. We only send and receive email on the basis of the terms set out at