Binary classification for ordinal regression
rossdiener opened this issue · comments
This paper talks about the benefit of using binary classification for ordinal regression: https://papers.nips.cc/paper/3125-ordinal-regression-by-extended-binary-classification.pdf
The major benefits are: 1) no need to invent new loss function and optimization; 2) whatever optimization goes to Logistic Regression, the implementation gets it for free.
- Microsoft uses this implementation:
https://docs.microsoft.com/en-us/azure/machine-learning/studio-module-reference/ordinal-regression - Amazon also uses this implementation: https://www.researchgate.net/publication/318108212_Ranking_and_Calibrating_Click-Attributed_Purchases_in_Performance_Display_Advertising