How to change Adaptive Model technique from Bayesian to Gradient Boosting?

General
, , ,
1

Hi Team,

I would like to understand whether the Adaptive Model technique in Pega can be changed from Bayesian to Gradient Boosting.

  • Is the modeling technique selected automatically by Pega by default, or is there a configuration option to choose between Bayesian and Gradient Boosting?
  • For an existing Adaptive Model, can we switch the technique from Bayesian to Gradient Boosting, or do we need to create a new model?
  • If switching is supported, what is the recommended approach and are there any impacts on model history, learning, or predictors?

Any guidance or documentation references would be appreciated.

2

Hi @YashovardhanM1202 , when you create an adaptive model in Prediction Studio there is an option to choose the modelling technique. If you are using an existing Prediction you can also replace an adaptive model in the Prediction. If you are already using adaptive models and these are Bayesian, then you can use the ‘Add Challenger’ action in a Prediction. See Pegasystems Documentation

Regards,

Ivar

3

Hi Yashovardhan,

You can’t switch the technique on a live model, so the usual approach is Champion–Challenger configured in Prediction Studio:

  • Shadow Mode first, the new Gradient Boosting model learns alongside the Bayesian one but doesn’t affect decisions
  • Then Champion–Challenger, give it a small share of traffic once performance looks good
  • Finally, promote it fully once you’re confident

Low risk, and you get a proper like-for-like comparison on real traffic.

Andy