Orange Forum • View topic - GSoC idea: multilabel classification

GSoC idea: multilabel classification

General discussions about Orange and with Orange connected things (data mining, machine learning, bioinformatics...).

GSoC idea: multilabel classification

Postby dknight » Sun Mar 20, 2011 15:27


I am interested in working for the idea of multilabel classification as a GSoC project. I could not find any mentor listed in the ideas page. I would like to discuss the idea with the supposed mentor for the project. I am skilled in C and Python and also know Java. I worked on classification of faces using machine learning techniques using MATLAB.


Re: GSoC idea: multilabel classification

Postby Blaz » Mon Mar 21, 2011 15:33


Idea for this task was to develop a library in Python and Orange for multilabel classification. There are excellent libraries for these methods in Java, like mulan, and one possibility is its reimplementation, testing, and crafting of nice documentation with examples in Orange.


Re: GSoC idea: multilabel classification

Postby biran0079 » Wed Mar 23, 2011 5:38

I am also interested in this project.
I have experience in several machine learning research projects: hand-written digit classification, POS tagging using HMM, multi-label classification using conditional random field(CRF) based algorithms(currently working on, almost done).
I am confident in implementing multi-label machine learning algorithms(since I am working on it), and I am familiar with C,C++ and python(with C++ and python GUI programming experience).

Re: GSoC idea: multilabel classification

Postby Blaz » Wed Mar 23, 2011 10:13

Great. This could be a really interesting project. Especially since it is not only about the coding. Orange's libraries come with examples, data sets, and demonstrations, all to help user's to start with coding through re-use of the past examples.

Re: GSoC idea: multilabel classification

Postby daffes » Thu Mar 24, 2011 13:15


This project also got my attention. I have some years of experience with C++ and Python, and I also have a good level of Java to explore the library Mulan. My experience with Artificial Intelligence is not vast, as I did only two courses in the university with some projects. Besides that, I did many courses in Computer Theory and I'm very used to read papers and implement difficult algorithms and data structures aiming Programming Contests, and I'm willing to work hard to be quickly familiar with all the concepts needed.

Would this project be assigned for a single person?

Thank you,
Davi Costa.

Re: GSoC idea: multilabel classification

Postby Blaz » Thu Mar 24, 2011 13:22

Dear Davi, probably yes (to a single person). It of course all depends how many students will be assigned to Orange by Google, but multi-label classification would on its own be better fitted to one student only.

Re: GSoC idea: multilabel classification

Postby Mitar » Thu Mar 24, 2011 16:26

It is a Google requirement that only single student is doing a given project. I can wish you only that you make good proposals and good application. Also use this time before the application to discuss with us details so that proposal will show you already understand what we want from you. If you invest time now, this is already a bonus. Maybe think about other ideas, come up with your own idea and propose it (you can of course discuss it here in advance, too). And do not be afraid of competition. This should only make you peform better.

Re: GSoC idea: multilabel classification

Postby hamzah » Fri Mar 25, 2011 15:10

I am working on Linear Discrimination Analysis for multilablel and SVM.
I wish to know how will you select students for this project?

Re: GSoC idea: multilabel classification

Postby matija » Fri Mar 25, 2011 16:12

Orange currently lacks any support for multi-label classification, so we would prefer an implementation of general framework (data manipulation, basic algorithms such as combinations of single-label classification methods, evaluation methods, ...) over sophisticated algorithms alone.

Re: GSoC idea: multilabel classification

Postby Mitar » Fri Mar 25, 2011 18:46

hamzah wrote:I wish to know how will you select students for this project?

We will be selecting them based on the proposal, which should be based on our template. There you can see what we are interesting in. And based on that we will select students. Along with that you have also convince us that you have enough technical/computer skills and knowledge to really do what you are proposing. So proposals should be realistic.

Until the deadline you have time to discuss your ideas with us. You should use that time. So that proposal will be as much as possible to what we want to hear and what we think it is needed or useful (or what you convince us, too). And as I said, also that you can do it. So references, code examples, your prior open-source work, all this counts.

Return to General Discussions