Here's something that'll create a JSON model that can be directly uploaded into Solr:
https://github.com/ryac/lambdamart-xml-to-json It'll map the feature IDs to the names found in the feature-store as well. I had this error when uploading model: Model type does not exist org.apache.solr.ltr.model.MultipleAdditiveTreesModel Once I changed the values from floats to strings it uploaded fine. Ryan On Mon, 24 Jul 2017 at 12:14 Ryan Yacyshyn <ryan.yacys...@gmail.com> wrote: > Thanks Doug, this is helpful. > > I also started something last night to output to JSON for Solr, I'll post > it up as well. > > Ryan > > > > > On Sun, 23 Jul 2017 at 23:48 Doug Turnbull < > dturnb...@opensourceconnections.com> wrote: > >> Yes you're correct that the feature is the 1-based identifier from your >> training data. >> >> For a script. Not one to Solr exactly, but when developing the >> Elasticsearch plugin, I started to work on a JSON serialization format, >> and >> as part of that built a Python script for reading the Ranklib XML and >> outputting to my own JSON format. It could be helpful to you or anyone >> constructing a script: >> >> >> https://github.com/o19s/elasticsearch-learning-to-rank/blob/7426858c2afb168ac426cab6d857fddccb9c26fc/demo/ranklibToJson.py >> >> On Sun, Jul 23, 2017 at 7:18 AM Ryan Yacyshyn <ryan.yacys...@gmail.com> >> wrote: >> >> > Hi everyone, >> > >> > I'm trying out the LTR plugin and have a couple questions when it comes >> to >> > converting the LambdaMart XML to JSON. Below is a snippet of the model >> > generated from rankLib: >> > >> > <ensemble> >> > <tree id="1" weight="0.1"> >> > <split> >> > <feature> 10 </feature> >> > <threshold> 0.28156844 </threshold> >> > <split pos="left"> >> > <feature> 11 </feature> >> > <threshold> 7.111111 </threshold> >> > <split pos="left"> >> > <feature> 7 </feature> >> > <threshold> 2.2759523 </threshold> >> > <split pos="left"> >> > <output> 0.8436763 </output> >> > </split> >> > <split pos="right"> >> > <output> 1.4320849 </output> >> > </split> >> > </split> >> > ---------------------- >> > >> > And a sample of training data: >> > >> > 1 qid:1 1:0.0 2:1.0 3:0.0 4:0.0 5:0.0 6:0.0 7:19.496738 8:0.0 9:0.0 >> > 10:0.08307255 11:7.111111 #docId: oeqzg5-165248 >> > >> > It's probably obvious, but I just want to check if the <feature> node in >> > the XML is referring to the ID in my training set and that it's possible >> > that the model doesn't use all features in the training data? >> > >> > I'll be mapping these feature IDs with the names I gave them in the >> feature >> > store in Solr.. >> > >> > Is there a script or utility already made out there to convert the XML >> to >> > this Solr JSON format? The closest I found to something was this: >> > https://sourceforge.net/p/lemur/feature-requests/144/ >> > >> > Thanks for your help! >> > >> > Ryan >> > >> >
