Multinomial Logistic Regression in Rapidminer

sunnyal

Hi,

 

I wanted to perform a Multinomial Logistic Regression for designating our customer types. I do not see any operator for this can you provide some guidance in this regards

 

Thx

 

Thomas_Ott

@sunnyal the Generalized Linear Model operator can do Multinominal labels, just have to set the Family parameter to multinominal

rfuentealba

Hi @sunnyal,

 

Is there any difference between polynomial and multinomial logistic regressions? Because if there isn't a difference, I remember that there is an operator named Logistic Regression (Evolutionary), that you can configure to do polynomial regressions.

 

All the best,

 

Thomas_Ott

@sunnyal the Generalized Linear Model operator can do Multinominal labels, just have to set the Family parameter to multinominal

sunnyal

Thank you Rodrigo, Tom,

 

I also see an operator called Polynomial regression. Would this suffice the need for performing multinomial regression?? Is there a difference between this an Logistic Regression (Evolutionary) and Generalized Linear Model ??

 

Also, do we have an sample process on Generalized Linear Model with family type as multinomial, that I can infer ??

 

Thx 

Thomas_Ott

@sunnyal The Polynominal Regression operator can only use a numerical label with numerical labels, so you can't use it for a multi-label application. The Logistic Regression (Evolutionary) operator is a lot like a standard LR algo BUT uses a Support Vector Machine to determine the boundary conditions of a binomal label. So that won't work either if you have multi-labels (i.e. more than 2). 

 

The best bet, IMHO, is to use the GLM operator. It's a modern implementation and really fast.