You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
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.
The text was updated successfully, but these errors were encountered:
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.
![Screenshot 2020-05-19 at 10 22 41 AM](https://user-images.githubusercontent.com/1763321/82277535-b4880000-99ba-11ea-8c62-98ebc18e3d46.png)
The text was updated successfully, but these errors were encountered: