Lambdarank does not work on MSLR-WEB10K

[zhihuacc]

Hi, I run the code on MSLR-WEB10K and found that ranknet/default and ranknet/factor can converge on MSLR-WEB10K. E.g., NDCG@5 can increase from 0.3 to higher than 0.4. However the lambdarank/default does not converge. The training loss increases from the begining.

[akanyaani]

Hi @harvey-che

Thanks for reporting the issue, let me generate it on my system.
I will resolve once I find the issue.

Thanks