limit range for some Regression coefficient

Karas

Hi,

I'm using Rapidminer to do some performance/properties prediction using for example linear regressions.

In my case some of the found regression coefficients are against the current knowledge.

Would there be a way to stipulate a certain range or sign for some of the regression coefficients (similar to "min" and "max coefficient" in the polynomial regression)

The idea would be to get a somewhat less accurate prediction model but correct from a physics point of view.

The only thing I could think about for the moment would be to use an "Optimize parameter" for the "known" attributes and let the linear regression operator calculate the others by using the Performance+"known attributes" as label

If someone has a better and more efficient idea, I'd be grateful

Thanks