[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index] [List Home]
Re: [recommenders-dev] About the license issue of the incubator project


Hi Marcel,


great to hear you want to continue the development of Precise at Eclipse! As I follow your project since a while now, I'd be pretty happy if we could support you and your  group's project here at Eclipse. 

Just a couple of questions out of curiosity and to learn a bit more about your ideas:
Is this project also backed up by your research group or mostly developed by you? As you have almost finished your PhD (hopefully ;)) what are your future plans with Precise? Will it be continued by you even with a new job or will your research group continue the work on it? 


The existing code was originally developed by two students and me, then improved by another three students in my group. 
Since the whole project is not that large, I think I know almost all the code well.
Because I am personally interested in this project and Eclipse Code Recommenders, I will certainly keep doing it even after I get a real job ;)
As for my research group, I am not totally sure how many people will join in in the long term. For now I am likely to get one or two excellent 
guys to help.
 

Answer to #1:
The Apache License of Mahout is fine. The only issue with Mahout is that it has a huge set of dependencies to other libraries (the distribution is 65MB large) which may need to be IP checked before it can be shipped from eclipse.org. Do you need Mahout for generating models only or also at runtime to compute proposals? 
BTW: What algorithms are you using?

Currently we are using k-NN search, so Weka (or Mahout) is just used for computing proposals.
As k-NN is an instance-based scheme, we do not have to build models using Weka (or Mahout).
 

To answer #2:
We have our own implementation of Bayesian Networks to compute recommendations which is at eclipse.org and free to reuse.
The models are generated by some self-written code which will move to o.e.r.rdk soon. Early versions are in the o.e.r.mining but it's very specific to our call recommender and not very reusable for other usages. But we'd like to use Mahout for clustering code search results in the near future. So we should strive for having Mahourt

Let me ask Wayne how to get Mahout (or a subset of it) through IP check.
 
Marcel


Great to hear that. Then we are planning to replace Weka with Mahout :)
Thank you very much for your help!


Cheng
 


On 18.06.2012, at 18:03, Cheng Zhang wrote:

Hi,

I am Cheng Zhang, a PhD student from Shanghai Jiao Tong University, who is very interested in the Eclipse Code Recommenders project. 

My colleagues and I have developed a prototype of a parameter guessing engine, named Precise, and now we are trying to make it an Ecilpse Recommenders Incubator project as described in: 

The project is current maintained by two undergraduates and the code still needs refinement.
But in the first place, I have a problem with the license: our project is using Weka as the machine learning engine, but Weka is under GPL license.

It seems that we have to replace Weka with other machine learning tools or implement the algorithms by ourselves. So far we have come up with two possible solutions:
1. Use Apache Mahout. Mahout is under Apache License, Version 2.0. Since we have little knowledge about license issues, we are not sure whether this license is permissible in the incubator project or not. Could anyone give some guidance?
2. Reuse the machine learning component in Eclipse Code Recommenders. Also I am wondering whether this solution is feasible.

Any comment or suggestion will be appreciated. Thank you!

Best,
Cheng
_______________________________________________
recommenders-dev mailing list
recommenders-dev@xxxxxxxxxxx
http://dev.eclipse.org/mailman/listinfo/recommenders-dev

-- 


_______________________________________________
recommenders-dev mailing list
recommenders-dev@xxxxxxxxxxx
http://dev.eclipse.org/mailman/listinfo/recommenders-dev