I've been doing research on feature selection, but I'm missing some codes that are available to help me, and I hope that someone can help, provide a feature selection method based on UCI data classification. Thank you!
Do you mean that you need a masters thesis with that title, because the scope is general from your title. I think you would find what you are looking for if you are more specific
The way I would go about it is the following. As you have been doing research about this subject, you certainly found more than a few super relevant papers about it. The best way is simply to write a courteous message to the researchers that wrote these papers having the following ingredients:
1) Explain your interest in their paper (why it is relevant to your thesis)
2) Ask for the possibility to share their code to help you in your research. You can also ask them for advice about how the go about this particular software problem. Maybe there is an open source library doing what you specifically need that you are not aware of. I usually propose to send them corrections or bug descriptions if you find some.
3) If they answer favorably, thank them properly. Don't forget to thank them in your thesis. Send them a copy once you finish is also a nice gesture.
If you do this properly, it could be the start of your research network with people doing good stuff in your field...
Best of luck with this "meta-advice",
Bruno Martin
Dear Zaji
you can use the production of Faradars Company. please see the http://faradars.org/. in this web site all the production is produced numerous of production for feature selection.
best regard
Dear Ahmed,
I understand that you evaluate/create algorithms that select the most useful features - out of pool of existing features (raw or built) of a given dataset - in order to perform a certain task with a given algorithm. A lot has been published in this context, and most are constructs on top of information theory and sensitivity analysis. If you have no done so, I would invite you do dive into these concepts and publish your own code on OSF (http://osf.io).
If you are looking for specific code that is at the core of specific papers, then Bruno Martin's advice is very sound. You will need luck, otherwise you may have to use my first paragraph as plan B ;)
If you do have to dive into the two topics highlighted above, there's one potential pitfall you must be aware of: entropy. The information theory concept was created for a very specific problem and has been stretched way beyond its original scope. I haven't found a mathematical treatment of these extensions that makes me feel comfortable with using it for non-binary or non-streamed data, or both - you will find papers that share this perspective.
Regards
Perhaps you can try contacting a Dr Osama Alia from Tabuk University. He did his PhD in this area (I think)... forget his email tho.
Hi
I have code and a paper for you:"infinite feature selection". This technique is able to deal with hige amount of features and it will perform the rank of your variables / features. Then you can select a subset starting from the outcome of the method. It ranks the features by importance. The concept of importance can be changed by editing the basic function used for weight the graph. The method is unsupervised. The source code for matlab is available on the project page
Hi,
here is the source code of Infinite Feature Selection:
http://www.mathworks.com/matlabcentral/fileexchange/54763-infinite-feature-selection
here is the paper:
https://www.researchgate.net/publication/282576688_Infinite_Feature_Selection
best wishes,
you can check the references of this paper:
http://www.sciencedirect.com/science/article/pii/S0031320316300954
- Badges
- Science topic