I'd suggest that a good thing to look at would be functional tests of the metrics and algorithms in Hanxiao Sun's work from this summer. You'll generally need to go either to the original paper, or find an alternative implementation, to build up a series of tests that demonstrate that the implementation is doing what it is supposed to. Xapian-core contains a test framework which it would be sensible to reuse for this. (Tests can be fairly easily merged at a later point because they'll be largely independent of other tidying up needed to merge this summer's work.) J> On 1 Dec 2014, at 04:47, Olly Betts <olly at survex.com> wrote: > >> On Sun, Nov 30, 2014 at 02:09:50PM -0500, Manu Gupta wrote: >> I will try to work on : >> http://trac.xapian.org/wiki/GSoCProjectIdeas#Project:LearningtoRank >> >> I will be taking a Machine Learning class the next semester and I hope that >> this project will help me supplement my learning in Machine Learning and >> also gain a bit of knowledge in IR. >> >> If you can give me ideas on how to get around with the code for LTR >> project, it will be awesome. I can look at it and get back to you. > > Unfortunately right now there's not a single place with the code you > need to look at. > > Parth's work from GSoC 2011 is on master in the xapian-letor > subdirectory, and James has gone through the 2012 work to pick out the > useful changes, and I merged those to master earlier today. > > But in GSoC 2014 we had 2 letor projects working on different aspects, > and we need to sort out getting those both cleaned up and merged. We > should manage to do that before GSoC 2015, but right now you probably > need to look at both those branches: > > http://trac.xapian.org/wiki/GSoC2014/Learning%20to%20Rank%20Hanxiao%20Sun > http://trac.xapian.org/wiki/GSoC2014/Learning%20to%20Rank%20Jiarong%20Wei > > Cheers, > Olly > > _______________________________________________ > Xapian-devel mailing list > Xapian-devel at lists.xapian.org > http://lists.xapian.org/mailman/listinfo/xapian-devel
I have merged Jiarong's branch and xapian/master in my master brach: https://github.com/parthg/xapian It builds (you would have to .nobootstrap xapain-bindings). It would be also good to test the feature calculation modules in the new API framework (the above branch) if you are more interested in the general framework of the LTR. But if you are more keen on the ML part then Hanxiao's branch would be a good point to start and to do what James is suggesting. On Mon, Dec 1, 2014 at 8:26 AM, James Aylett <james-xapian at tartarus.org> wrote:> I'd suggest that a good thing to look at would be functional tests of the > metrics and algorithms in Hanxiao Sun's work from this summer. You'll > generally need to go either to the original paper, or find an alternative > implementation, to build up a series of tests that demonstrate that the > implementation is doing what it is supposed to. > > Xapian-core contains a test framework which it would be sensible to reuse > for this. > > (Tests can be fairly easily merged at a later point because they'll be > largely independent of other tidying up needed to merge this summer's work.) > > J > > > On 1 Dec 2014, at 04:47, Olly Betts <olly at survex.com> wrote: > > > >> On Sun, Nov 30, 2014 at 02:09:50PM -0500, Manu Gupta wrote: > >> I will try to work on : > >> http://trac.xapian.org/wiki/GSoCProjectIdeas#Project:LearningtoRank > >> > >> I will be taking a Machine Learning class the next semester and I hope > that > >> this project will help me supplement my learning in Machine Learning and > >> also gain a bit of knowledge in IR. > >> > >> If you can give me ideas on how to get around with the code for LTR > >> project, it will be awesome. I can look at it and get back to you. > > > > Unfortunately right now there's not a single place with the code you > > need to look at. > > > > Parth's work from GSoC 2011 is on master in the xapian-letor > > subdirectory, and James has gone through the 2012 work to pick out the > > useful changes, and I merged those to master earlier today. > > > > But in GSoC 2014 we had 2 letor projects working on different aspects, > > and we need to sort out getting those both cleaned up and merged. We > > should manage to do that before GSoC 2015, but right now you probably > > need to look at both those branches: > > > > > http://trac.xapian.org/wiki/GSoC2014/Learning%20to%20Rank%20Hanxiao%20Sun > > > http://trac.xapian.org/wiki/GSoC2014/Learning%20to%20Rank%20Jiarong%20Wei > > > > Cheers, > > Olly > > > > _______________________________________________ > > Xapian-devel mailing list > > Xapian-devel at lists.xapian.org > > http://lists.xapian.org/mailman/listinfo/xapian-devel > > _______________________________________________ > Xapian-devel mailing list > Xapian-devel at lists.xapian.org > http://lists.xapian.org/mailman/listinfo/xapian-devel >-------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.xapian.org/pipermail/xapian-devel/attachments/20141203/df4b5fb9/attachment-0002.html>
Hi Maybe I can start off with Hanxiao's branch but I can see there are two repos. https://github.com/v-hasu/xapian/tree/gsoc2014-evaluation https://github.com/v-hasu/xapian/tree/gsoc2014 which one should I pick? I have already successfully built xapian from the git repository ( https://github.com/xapian/xapian) Since I am new to ML (well I dont know anything about it?), should I read the paper try to make sense out of it and get back too? Best Regards Manu Gupta On Wed, Dec 3, 2014 at 5:08 AM, Parth Gupta <pargup8 at gmail.com> wrote:> I have merged Jiarong's branch and xapian/master in my master brach: > https://github.com/parthg/xapian > > It builds (you would have to .nobootstrap xapain-bindings). > > It would be also good to test the feature calculation modules in the new > API framework (the above branch) if you are more interested in the general > framework of the LTR. But if you are more keen on the ML part then > Hanxiao's branch would be a good point to start and to do what James is > suggesting. > > On Mon, Dec 1, 2014 at 8:26 AM, James Aylett <james-xapian at tartarus.org> > wrote: > >> I'd suggest that a good thing to look at would be functional tests of the >> metrics and algorithms in Hanxiao Sun's work from this summer. You'll >> generally need to go either to the original paper, or find an alternative >> implementation, to build up a series of tests that demonstrate that the >> implementation is doing what it is supposed to. >> >> Xapian-core contains a test framework which it would be sensible to reuse >> for this. >> >> (Tests can be fairly easily merged at a later point because they'll be >> largely independent of other tidying up needed to merge this summer's work.) >> >> J >> >> > On 1 Dec 2014, at 04:47, Olly Betts <olly at survex.com> wrote: >> > >> >> On Sun, Nov 30, 2014 at 02:09:50PM -0500, Manu Gupta wrote: >> >> I will try to work on : >> >> http://trac.xapian.org/wiki/GSoCProjectIdeas#Project:LearningtoRank >> >> >> >> I will be taking a Machine Learning class the next semester and I hope >> that >> >> this project will help me supplement my learning in Machine Learning >> and >> >> also gain a bit of knowledge in IR. >> >> >> >> If you can give me ideas on how to get around with the code for LTR >> >> project, it will be awesome. I can look at it and get back to you. >> > >> > Unfortunately right now there's not a single place with the code you >> > need to look at. >> > >> > Parth's work from GSoC 2011 is on master in the xapian-letor >> > subdirectory, and James has gone through the 2012 work to pick out the >> > useful changes, and I merged those to master earlier today. >> > >> > But in GSoC 2014 we had 2 letor projects working on different aspects, >> > and we need to sort out getting those both cleaned up and merged. We >> > should manage to do that before GSoC 2015, but right now you probably >> > need to look at both those branches: >> > >> > >> http://trac.xapian.org/wiki/GSoC2014/Learning%20to%20Rank%20Hanxiao%20Sun >> > >> http://trac.xapian.org/wiki/GSoC2014/Learning%20to%20Rank%20Jiarong%20Wei >> > >> > Cheers, >> > Olly >> > >> > _______________________________________________ >> > Xapian-devel mailing list >> > Xapian-devel at lists.xapian.org >> > http://lists.xapian.org/mailman/listinfo/xapian-devel >> >> _______________________________________________ >> Xapian-devel mailing list >> Xapian-devel at lists.xapian.org >> http://lists.xapian.org/mailman/listinfo/xapian-devel >> > >-------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.xapian.org/pipermail/xapian-devel/attachments/20141204/8253d403/attachment-0002.html>