need clarification in documentation

[leo-gan]

in https://github.com/Microsoft/LightGBM/blob/master/docs/Parameters.md

• "fair, Fair loss" - link here is broken and didn't point to the explanation;
• "fair_c, default=1.0, type=double ... parameter for Fair loss." - as previous
• "lambdarank, lambdarank application" - there is no link to the explanation. Google didn't bring anything meaningful;
• "goss, Gradient-based One-Side Sampling" - there is no link to the explanation. Google didn't bring anything meaningful;

also in Metric Parameters:

• "fair, Fair loss" - as previous;
• "auc, AUC" - link is wrong, it should be [https://en.wikipedia.org/wiki/Receiver_operating_characteristic#Area_under_the_curve]
• "binary_logloss, log loss" - link to the empty page

[wxchan]

lambdarank: https://pdfs.semanticscholar.org/fc9a/e09f9ced555558fdf1e997c0a5411fb51f15.pdf

[Laurae2]

For simple understanding:

• Fair loss: derivable proxy for Mean Absolute Error (MAE) objective function, check https://www.kaggle.com/c/allstate-claims-severity/discussion/24520 for more details about MAE proxies
• Fair loss c parameter: check https://www.kaggle.com/c/allstate-claims-severity/discussion/24520 - bigger c induces more brutal transitions (closer to V curve), while smaller c induces a smoother transition (closer to U curve) between the expected proxy MAE and the real MAE (U curve vs V curve). Real MAE is V curve, proxy MAE is U curve (similar to parabolas like Mean Squared Error). Read formula on the link to understand how it can change.
• LambdaRank: objective function for ranking tasks (its superset algorithm is LambdaMART - you might find more information about it on Google)
• Gradient-based One-Side Sampling: an alternative boosting method which seems to be performing Gradient sampling (according to the code), thus it is an adaptive method exploiting information (N.B: as it is an adaptive method, you cannot use bagging)

[henry0312]

I'm sorry for my late response.
However, I can't find another documetns for fair loss 😢

[wxchan]

the loss function name is really unfriendly for search engine, and seem this loss function is not mentioned a lot too

[github-actions]

This issue has been automatically locked since there has not been any recent activity since it was closed. To start a new related discussion, open a new issue at https://github.com/microsoft/LightGBM/issues including a reference to this.