Implementation of the Gamma distribution about projpred HOT 7 CLOSED

Hi,

I'm on vacation currently until July 27th, so I won't be able to provide a fast solution. Nonetheless, I can tell you that finishing support for gamma distributions is on our roadmap but not top priority right now. I will, however, take a close look at this and report back on the status.

Thanks for the interest and feedback!

from projpred.

That's no problem at all. I just wanted to decide if it is worth waiting with the analyses until the Gamma distribution is implemented or not, but I have no urgent problems with the current implementation (the models look fine).

I'll try the Gamma distribution later then, as soon as it is available (I have other analyses also with Gamma distributed dependent variables).

Thanks for the reply!

from projpred.

Hi,

I wondered whether there is any update on the implementation of the gamma distribution in projpred. Wanting to use it for a current project.

Thanks for all the great work!

from projpred.

Hi,

Thanks for your interest. We are not implementing specific families per se but a feature under development would support many models and should be readily available soon.

Basically, we are bringing support to all models of the form:

y ~ p(g(\eta)) 
\eta ~ p(\eta | X, Z, ..)

where g is a link inverse function and the function of \eta can be a linear or additive multilevel model. Does this clarify which kind of models will be supported?

Edited: confusion on the previous answer.

from projpred.

Hi,

Thanks for you quick response. Sorry it is not completely clear to me yet, I'm quite the novice at this. I thought the Gamma distribution was part of the exponential family, you seem to imply it is not. Can p() be any function? Does a regression using a Gamma distributed dependent variable with a log link fall under the form you specify?

from projpred.

I am sorry I didn't think twice about my answer!!

The answer is valid from a different angle: we are not going to implement family specific solutions but this approach will work as well for families not implemented that belong in the exponential family.

Sorry about the confusion.

from projpred.

Ah great, that is clear now :) Thanks for the response and will be looking out for that update!

from projpred.