Hi, you will have to be a little bit more specific, if anyone should answer your question. A good starting point would be the operator(s) names. Narrowing the question on the field of interest could be helpful, too...
In each case you have a 1/0 outcome variable. In the logit case, you do not assume that the observations are related; in the conditional logit, you assume they are. For example, say you have a family which is making a decision between 4 different options. Clearly, if they choose option 1, that is not independent of whether or not they choose option 2. In that case, using a logit would be inappropriate because it would not take into account the dependence between the observations.
Personally, I am looking for a way to do a conditional logit in RapidMiner. It does not seem to be possible. I have done it before in Stata and Matlab. Does anyone have any ideas?
Addendum: A hierarchical Bayes model would also work. I believe they are asymptotically equivalent. Unfortunately, I also can't figure out how to make a hierarchical Bayesian model....
Hi Eva, unfortunately we don't have a conditional logit learner in RapidMiner, yet, but I don't know if WEKA provides one. Then you simply could use it directly from RapidMiner. By the way: There are several model included in RapidMiner, which will model a conditional distribution implicitly like the decision tree.
I don't know exactly if this helps, but you could construct new, (possibly additional) attributes, containing the combination of the old attributes. This would somehow reflect some sort of dependency, but models it in a quite different way...
Contributor I
Answers
you will have to be a little bit more specific, if anyone should answer your question. A good starting point would be the operator(s) names.
Narrowing the question on the field of interest could be helpful, too...
Greetings,
Sebastian
Personally, I am looking for a way to do a conditional logit in RapidMiner. It does not seem to be possible. I have done it before in Stata and Matlab. Does anyone have any ideas?
Addendum: A hierarchical Bayes model would also work. I believe they are asymptotically equivalent. Unfortunately, I also can't figure out how to make a hierarchical Bayesian model....
unfortunately we don't have a conditional logit learner in RapidMiner, yet, but I don't know if WEKA provides one. Then you simply could use it directly from RapidMiner.
By the way: There are several model included in RapidMiner, which will model a conditional distribution implicitly like the decision tree.
I don't know exactly if this helps, but you could construct new, (possibly additional) attributes, containing the combination of the old attributes. This would somehow reflect some sort of dependency, but models it in a quite different way...
Greetings,
Sebastian