Fix irreproducible inference due to the imprecise floating point values in xgboost's get_dump

o19s / elasticsearch-learning-to-rank

Plugin to integrate Learning to Rank (aka machine learning for better relevance) with Elasticsearch

http://opensourceconnections.com/blog/2017/02/14/elasticsearch-learning-to-rank/

Apache License 2.0

1.48k stars 370 forks source link

Fix irreproducible inference due to the imprecise floating point values in xgboost's get_dump #500

Open pltb opened 2 months ago

pltb commented 2 months ago

Tries to address the issues of using the output of the get_dump method in xgboost as the import format by parsing the output of save_model instead

More context here https://github.com/o19s/elasticsearch-learning-to-rank/issues/497

TODO:

[x] parse using ObjectParser
[ ] add more tests
[ ] review from the plugin's authors

wrigleyDan commented 1 day ago

Summary of our chat at Haystack:

Rename the parser to model/xgboost+json+raw
Add the switch-case for the additional model parser
Preserve existing functionality
Add tests
Documentation (Daniel)
Add notebook to hello-ltr Github repo (Daniel)

pltb commented 1 day ago

@wrigleyDan a small correction: shall we rename it to model/xgboost+json+raw instead of v2?

wrigleyDan commented 12 hours ago

@wrigleyDan a small correction: shall we rename it to model/xgboost+json+raw instead of v2?

Thanks, copied the old name and forgot to change it. Well spotted, corrected it in the comment above.