On 25 October 2015 at 19:44, olologin <ololo...@gmail.com> wrote: > On 10/25/2015 08:12 PM, Raphael C wrote: >> >> From my quick reading of the thread it seems that people aren't >> convinced LambdaMART is very good in practice. Is that right/wrong? >> >> Raphael >> > > http://research.microsoft.com/en-us/um/people/cburges/tech_reports/MSR-TR-2010-82.pdf > says that: > >LambdaMART is the boosted tree version of LambdaRank, which is based > on RankNet. RankNet, LambdaRank, and LambdaMART have proven to be very > successful algorithms for solving real world ranking problems: for > example an ensemble of LambdaMART rankers won Track 1 of the 2010 Yahoo! > Learning To Rank Challenge. >
I am absolutely not an expert on this topic so please bear that in mind. I was merely summarising the discussion on the Pull Request as I saw it. Having said that, the issue seems to be a comparison with gradient boosted trees (GradientBoostingRegressor). The scikit-learn devs like to have a public dataset where a new technique does better than an already implemented one I believe before accepting an addition. It would be great if an expert could comment on this. Raphael ------------------------------------------------------------------------------ _______________________________________________ Scikit-learn-general mailing list Scikit-learn-general@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
