"Linear regression, removing the constant term from the model"

d1m0s

Hi guys,
I'm trying to run a linear regression model with dummy variables and I need to get rid of the constant term in my model but I do not see how to do this. It seems it can not be done through the Rapidminer's GUI (I am using learner\supervised\functions\linearregression operator).
Please help me.

[attachment deleted by admin]

IngoRM

Hello,

I am afraid this is currently not possible. I must admit that I never had a problem where this would have made any sense though 

Cheers,
Ingo

keith

Just offering my opinion...

If I understand the question correctly, he's essentially looking for the equivalent to the "Constant is Zero" option in Excel's Regression functionality in the Data Analysis add-in.

I haven't had to do it with RapidMiner yet, but I certainly have had circumstances where I wanted to model something with no constant. i.e. Structurally, I knew that if all the input variables were zero, the dependent variable would also have to be zero, and I wanted the regression to reflect that behavior, while otherwise using regression to find the best constants for the input variables.  So I could definitely see myself using such an option again in the future.

IngoRM

Hi again,

ok, if there are already two people out there who would like this, we have reached a critical mass 

Just kidding: Thanks for pointing this out again. II just have realized an example where this indeed could be very useful. So we have added a new parameter to the linear regression learner called "use_bias" which can be turned off. You can obtain the new version via CVS (described here: http://rapid-i.com/content/view/25/48/ ; please use the branch "Zaniah" for the latest developer version). Of course, this new feature will also be part of the next Enterprise Edition update.

Cheers,
Ingo

d1m0s

Wow...thanks a lot guys