Hi all, I am currently trying to compare different plant occurrence prediction maps generated in R and exported into GRASS. One of these maps was generated from a glm fitted to some data, and subsequently applying this glm model to a wider region using predict.glm. The outcome here was a probability of occurrence. The second map I generated using a gam (mgcv), however, this map seems to have assigned something like a negative log-likelihood of occurrence to each raster cell in the region. Since I would like to compare the two, I would like to figure out a way of having the same kind of output from the "predict" functions (either probability OR negative log-likelihood). Does anyone know of a way of changing the output options? And if not, does anyone have any suggestions of how I could deal with this issue? Thank you!
On Jun 18, 2009, at 3:23 PM, Anke Konrad wrote:> Hi all, > > I am currently trying to compare different plant occurrence > prediction maps generated in R and exported into GRASS. One of these > maps was generated from a glm fitted to some data, and subsequently > applying this glm model to a wider region using predict.glm. The > outcome here was a probability of occurrence. The second map I > generated using a gam (mgcv), however, this map seems to have > assigned something like a negative log-likelihood of occurrence to > each raster cell in the region. Since I would like to compare the > two, I would like to figure out a way of having the same kind of > output from the "predict" functions (either probability OR negative > log-likelihood). Does anyone know of a way of changing the output > options? And if not, does anyone have any suggestions of how I could > deal with this issue?Have you actually read the help pages for predict.glm and mgcv::predict.gam? -- David Winsemius, MD Heritage Laboratories West Hartford, CT
Anke, mgcv:predict.gam certainly didn't produce `something like a negative log-likelihood of occurrence', but is it possible that one of your maps is on the probability scale and the other on the linear predictor scale? If you used predict.glm(model1,type="response"), but predict.gam(model2,type="link"), then you'd get the sort of difference that you are maybe describing. ?predict.gam and ?predict.glm give more details. If that doesn't resolve the issue, then a few more details about the models actually being fitted are probably needed. best, Simon Quoting Anke Konrad <akonrad at nmsu.edu>:> Hi all, > > I am currently trying to compare different plant occurrence prediction > maps generated in R and exported into GRASS. One of these maps was > generated from a glm fitted to some data, and subsequently applying > this glm model to a wider region using predict.glm. The outcome here > was a probability of occurrence. The second map I generated using a gam > (mgcv), however, this map seems to have assigned something like a > negative log-likelihood of occurrence to each raster cell in the > region. Since I would like to compare the two, I would like to figure > out a way of having the same kind of output from the "predict" > functions (either probability OR negative log-likelihood). Does anyone > know of a way of changing the output options? And if not, does anyone > have any suggestions of how I could deal with this issue? > > Thank you! > > ______________________________________________ > R-help at r-project.org mailing list > https://stat.ethz.ch/mailman/listinfo/r-help > PLEASE do read the posting guide http://www.R-project.org/posting-guide.html > and provide commented, minimal, self-contained, reproducible code.
On Thu, 2009-06-18 at 13:23 -0600, Anke Konrad wrote:> Hi all, > > I am currently trying to compare different plant occurrence prediction > maps generated in R and exported into GRASS. One of these maps was > generated from a glm fitted to some data, and subsequently applying this > glm model to a wider region using predict.glm. The outcome here was a > probability of occurrence. The second map I generated using a gam > (mgcv), however, this map seems to have assigned something like a > negative log-likelihood of occurrence to each raster cell in the region. > Since I would like to compare the two, I would like to figure out a way > of having the same kind of output from the "predict" functions (either > probability OR negative log-likelihood). Does anyone know of a way of > changing the output options? And if not, does anyone have any > suggestions of how I could deal with this issue?Did you set type = "response" in both predict.glm and predict.gam? It sounds like the gam predictions are still on the scale of the link function... G> > Thank you! > > ______________________________________________ > R-help at r-project.org mailing list > https://stat.ethz.ch/mailman/listinfo/r-help > PLEASE do read the posting guide http://www.R-project.org/posting-guide.html > and provide commented, minimal, self-contained, reproducible code.-- %~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~% Dr. Gavin Simpson [t] +44 (0)20 7679 0522 ECRC, UCL Geography, [f] +44 (0)20 7679 0565 Pearson Building, [e] gavin.simpsonATNOSPAMucl.ac.uk Gower Street, London [w] http://www.ucl.ac.uk/~ucfagls/ UK. WC1E 6BT. [w] http://www.freshwaters.org.uk %~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~% -------------- next part -------------- A non-text attachment was scrubbed... Name: not available Type: application/pgp-signature Size: 197 bytes Desc: This is a digitally signed message part URL: <https://stat.ethz.ch/pipermail/r-help/attachments/20090619/f0aca008/attachment-0002.bin>