1

I'd like to use Catboost for actuarial models (eg claims frequency). Although I see that Poisson loss is an option, I don't see that exposures are directly supported. How do people deal with this?

shadowtalker
  • 12,551
thecity2
  • 1,955
  • Does Catboost provide a way to supply arbitrary custom instance weights? If so, you can work around the lack of offset/exposure support: https://stats.stackexchange.com/a/270151/36229. Otherwise you might be stuck making a feature request. it seems like XGBoost and LightGBM do appear to support fully custom objective functions, if you need an alternative. – shadowtalker Oct 31 '23 at 20:21
  • @shadowtalker I was recommended elsewhere to look into this functionality: https://catboost.ai/en/docs/concepts/python-reference_catboostclassifier_set_scale_and_bias – thecity2 Nov 02 '23 at 17:05
  • Seems reasonable, but it looks like you can only set a scalar scale, not a per-observation scale to act as the exposure. – shadowtalker Nov 02 '23 at 20:22
  • @shadowtalker hmmm that's a problem then! Thanks for looking at it. – thecity2 Nov 03 '23 at 17:01

0 Answers0