Open Gu-Youngfeng opened 4 years ago
After tried many times to tune the parameters, I found that if I change the -metric2t ERR@10
to -metric2t NDCG@10
, the above error will be clear. However, the re-ranked list is the same with the original list. That is, AdaRank CANNOT change the relative order of configurations in the original list.
Maybe all the metrics to be optimized in a learning-to-rank model should be set to -metric2t NDCG@k
. That is, the command 1 should be,
>> java -jar RankLib.jar -train train_set.txt -ranker XX -metric NDCG@10 -save trainer.txt
Last commit a62edaf solved the issue #4 successfully, but it simultaneously triggered another error when running
reconfig.py
using ranker AdaRank, look at the following failing trace,There is a possibility that some parameters in AdaRank are mis-assignmented. Here are the commands we used to invoke ranker AdaRank. The error happened when running command 2.
Command 1: Training a model
trainer.txt
based ontrainset_txt15.txt
.Command 2: Re-ranking the
testset_txt15.txt
based ontrainer.txt
, then save the results inresult.txt