Use mean loss instead of sum loss in Game [backward incompatible] #252

fastier-li · 2017-03-13T16:23:44Z

While comparing spark.ml and scikit, we realized that Game is using a sum in the loss function computation, but that is not desirable for scaling and numerical stability. We should change our loss function calculations to use a mean instead.

fastier-li changed the title ~~Use mean loss instead of sum loss in Game~~ Use mean loss instead of sum loss in Game [backward incompatible] Mar 13, 2017

fastier-li self-assigned this Mar 15, 2017

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Use mean loss instead of sum loss in Game [backward incompatible] #252

Use mean loss instead of sum loss in Game [backward incompatible] #252

fastier-li commented Mar 13, 2017 •

edited

Loading

Use mean loss instead of sum loss in Game [backward incompatible] #252

Use mean loss instead of sum loss in Game [backward incompatible] #252

Comments

fastier-li commented Mar 13, 2017 • edited Loading

fastier-li commented Mar 13, 2017 •

edited

Loading