benwtrent commented on issue #13611: URL: https://github.com/apache/lucene/issues/13611#issuecomment-2253065836
OK, after correcting my testing function, here are the `k`s I had to set to get the true NN. ``` k=140 with graph settings m=16 and b=100 k=145 with graph settings m=16 and b=150 k=145 with graph settings m=16 and b=200 k=5 with graph settings m=16 and b=250 // Seems like the sweet spot k=165 with graph settings m=32 and b=100 k=165 with graph settings m=32 and b=150 k=165 with graph settings m=32 and b=200 k=170 with graph settings m=32 and b=250 k=170 with graph settings m=48 and b=100 k=170 with graph settings m=48 and b=150 k=170 with graph settings m=48 and b=200 k=170 with graph settings m=48 and b=250 k=170 with graph settings m=64 and b=100 k=170 with graph settings m=64 and b=150 k=170 with graph settings m=64 and b=200 k=170 with graph settings m=64 and b=250 ``` It seems given the order of your data, we are hitting local minima pretty often. Increasing the beamWidth for your format seems like the best option as it will increase graph quality, allowing the graph building itself to not get stuck in local minma. Or, you can set `k` higher. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org --------------------------------------------------------------------- To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For additional commands, e-mail: issues-h...@lucene.apache.org
