Changing predictor type effectively means removing and adding a predictor. There is no way to retain previous responses - because Pega does not keep a record of all individual response data but instead only keeps binned aggregates.

Usually removing/changing a single predictor has little impact as models learn quickly given the response rates in typical applications. Also the impact of a single predictor is usually limited because there is often a large set of predictors, many of which show some correlations.

If you just want to study the impact of a different predictor type (or setting) you could also consider introducing a different adaptive model rule and plug that in into your strategy just before the current one - so it piggybacks on learning but its propensities are not used.