Recent changes to bugs-and-support

Infinite loop in KMeans

Roman Bliznets — Tue, 29 Jan 2013 12:01:20 -0000

I used JavaML 0.1.7
Some datas cause infinite loop during a clusterization with KMeans. I cannot provide data to test the issue but I debugged and found what problem is.
There is a loop: while ((randomCentroids) || ((iterationCount < this.numberOfIterations) && (centroidsChanged)))

When iterationCount exeded this.numberOfIterations, the randomCentroids variable still can be true. This cause infine loop when countPosition array contains a zero (0) values, e.g. [0, 400, 0, 0].

FileHandler.loadDataset (Only take ANSI encoding)

mahdi — Thu, 01 Nov 2012 10:52:21 -0000

Hello
i use the filehandler method that specifies the class value index and try to import a CSV file,
but there is a constraint that input file encoding should be ANSI and does not work with UNICODE

feature subset selection

mahdi — Fri, 19 Oct 2012 06:50:56 -0000

hello dear all!
i need to review the source code of "greedy forward selection" and "pearson correlation", i will be thankful that i could access to this codes.
i would like to thank you for any help that you may render me regarding my request.

NullPointerException in Bagger

Anonymous — Tue, 11 Sep 2012 18:14:22 -0000

Random Forest's buildClasiffier method calls this Bagging constructor:
bagger = new Bagging(trees, rg);
which keeps SamplingMethod instance object to null. Then when bagger.buildClassifier is build calls DatasetTools.sample whith SamplingMethod object null.
DatasetTools.sample breaks with a NPE.

I've solved this with calling the Bagging constructor that receives the dataset, a sampling method and a random seed. The sampling method by default is new SubSampling().

Bug in net.sf.javaml.distance.EuclideanDistance

Kartik Kukreja — Mon, 16 Jul 2012 01:10:05 -0000

I think there's a typo in the calculateDistance(Instance x, Instance y) method. It throws an exception if the two instances have different number of attributes but checks this as follows : if (x.noAttributes() != x.noAttributes()). It should be : if (x.noAttributes() != y.noAttributes())

FileHandler.loadDataset and type String

Anonymous — Wed, 27 Jun 2012 09:15:42 -0000

FileHandler saves datasets with classValues of type String correctly. FileHandler.loadDataset doesn't load the classValue as String. The classValue is set to null.

OPTICS incorrect / incomplete

Anonymous — Tue, 05 Jun 2012 10:46:13 -0000

The implmentation of OPTICS seems to be incorrect: as far as I can tell, it will produce DBSCAN clusters, because of the way it handles the assignment of cluster IDs to the objects.

Your code seems to be copied from Weka, which received a *partial* OPTICS implementation by some random student. This is not the best place to start...

Please review the OPTICS implementation, and compare it to the publication.

ARFFHandler IndexOutOfBounds

Anonymous — Wed, 21 Mar 2012 17:09:33 -0000

ARFFHandler.java throws an index out of bounds when the class index is -1. Comparing with FileHandler.java, I believe the change is from this:

if (i > classIndex)
values[i - 1] = val;

To this

if (i != -1 && i > classIndex)
values[i - 1] = val;

LibSVM.getWeights()

Tran Tan Minh Dao — Sat, 07 Jan 2012 10:16:05 -0000

Dear Sirs,

I'm using Java-ml for my project which is a multi-class classification problem.

I successfully trained my data using Linear (kernel_type = svm_parameter.LINEAR) LibSVM, but I want to get the weights for each class
- which, as I understand, the parameters for my hypothesis.
I used the method LibSVM.getWeights().

As I have 20 classes and 400 features, I expected to get a double array (double[]) of the size 400x20 = 8000, but only got 400 instead.

I did many Google searches and get to know that LibSVM use one-against-one approach, but still I think there may be more weights than 400.
So I don't quite sure if this is a mistake or not. May be LibSVM discribed "weights" something I haven't understood, or maybe what I received
was only the weights for the 1st class, ... so I'm writing this and hope that someone could explain it to me, or give me some hints / links/ documents
for the problem.

I thank that someone in advance, and wish you all the best for the wonderful careers you are making.

A beginner of machine learning.

P.S: I also noticed the rawDecisionValues() method, but not quite had any idea what it means when you get a double array of these values.
I read some papers on SVM, but can't tell what-is-what or which is which from the things in the papers and the things in the API.

So if you please, enlight me about this too.

Sorry for my smattering and the time it took you for reading this.

how to create model

Ashwani Jha — Thu, 25 Aug 2011 10:57:32 -0000

I am using random forest for classification of my data, first it builds a classifier on training file (size ~1Gb) and then classifies the instances (testing file).
The problem is that i am developing a software, and generation of classifier every time to classify result data is not feasible.
I want to know that can i write this classifier to some model file, if so how can i use this model file to classify. Please help as i am a newbie.