o19s/elasticsearch-learning-to-rank

Does Ranklib2.16 contains the LR(Logistic Regression) model?

yzc1103 opened this issue · 1 comments

yzc1103 commented

According the link provided in the readme documentation of https://github.com/o19s/elasticsearch-learning-to-rank/tree/es_7_6_2, I downloaded ranklib2.16 and found that only 8 models were included(as described in the ranklib2.16 introduction documentation linked to https://sourceforge.net/projects/lemur/files/lemur/RankLib-2.16/). However, the 9th model LR was used in the current https://github.com/o19s/elasticsearch-learning-to-rank/tree/es_7_6_2/demo/train.py code.
So I'm confused about this situation. Can anyone answer that question?

nathancday commented

Linear regression (not Logistic regression) was supported in past versions of Ranklib. It looks like it was dropped sometime before 2.16, which came out earlier this year.

We have a hardened fork of Ranklib, called RankMcRankface which still has linear regression as a model option. This is the underlying tree building library the notebook code you referenced uses, instead of Ranklib directly from SourceForge.