[GRASS-user] r.regression.multi for logistic regression - it is possible?

[Nuno Sá]

Hey Roberto!

Never done it in Grass but its a typical operation.

For me it seems a bit confusing with for me since you are not showing me a
glm fitting but instead just using the occurrence map in the logistic link
function.


What you need to do is to fit a GLM to the binary data you have, something
you can do in R for example check here:

http://www.statmethods.net/advstats/glm.html


Then you will build your output in grass using the r.mapcalc: F(x) = 1 / (
1+ e^(-GLM)  (That will be your probability map).

But, you can actually do everything using R only, no need to come into
grass but you can if you want.

Good luck!
Ciao!


On 5 September 2014 10:46, Roberto Marzocchi <roberto.marzocchi at gmail.com>
wrote:

>  Dear all,
>
> I have dichotomous data (occurrence 1 and non-occurence 0) and I want to
> use the  logistic regression method in order to obtain susceptibility map
> of landslide occurrence.
>
> I try to use r.regression.multi, but I am not sure to correctly define the
> mapy variable..
>
> I try to use r.mapcalc
>
> *r.mapcalc
> expression="logit_occurrence=log(occurrence_map+0.00001)-log(1-occurrence_map+0.00001)"*
>
> obtaining a map that go from -11.5 (not occurrence) to +11.5 (occurrence)
>
> when I try to use r.regression.multi I obtain a lot of null values in
> residuals and estimates map.
>
> i do not understand why..
>
> Thanks in advanced for the attention
> Roberto
>
> _______________________________________________
> grass-user mailing list
> grass-user at lists.osgeo.org
> http://lists.osgeo.org/mailman/listinfo/grass-user
>



-- 

Nuno César de Sá
+351 91 961 90 37
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.osgeo.org/pipermail/grass-user/attachments/20140905/b7e284ca/attachment.html>