Error in applying visreg to fitted model

[mdmazur]

I have been using visreg for a while now to plot conditional effects with sdmTMB, and at the end of last week I have started getting this error:

Error in `predict()`:
! `xy_cols` (the column names for the x and y coordinates) are not in
  `newdata`.

It seems to run the model again when running the visreg function. I'm not sure why it seems to think the model should be updated, but I think that may be where the error is coming from. Any ideas as to why this is happening?

[Lewis-Barnett-NOAA]

Please make sure that the column names for your spatial coordinates are the same in the fitted model and your prediction data frame

On Tue, Sep 3, 2024 at 3:53 PM mdmazur @.***> wrote:

I have been using visreg for a while now to plot conditional effects with sdmTMB, and at the end of last week I have started getting this error:

Error in predict(): ! xy_cols (the column names for the x and y coordinates) are not in newdata.

It seems to run the model again when running the visreg function. I'm not sure why it seems to think the model should be updated, but I think that may be where the error is coming from. Any ideas as to why this is happening?

— Reply to this email directly, view it on GitHub https://github.com/pbs-assess/sdmTMB/issues/366, or unsubscribe https://github.com/notifications/unsubscribe-auth/AAKMJP4A7Q3PTXCUPKYRTCLZUY4WVAVCNFSM6AAAAABNTCXIG2VHI2DSMVQWIX3LMV43ASLTON2WKOZSGUYDGOJVGI3DSMI . You are receiving this because you are subscribed to this thread.Message ID: @.***>

-- Lewis Barnett, PhD (he/him) Research Fish Biologist

NOAA Fisheries, Alaska Fisheries Science Center 7600 Sand Point Way NE, Bldg 4 Seattle, Washington 98115 Google Voice: (206) 526-4111

[seananderson]

visreg does that part for you though. Can you provide a reproducible example, @mdmazur? You could start with this:

library(sdmTMB)
fit <- sdmTMB(
     density ~ s(depth_scaled),
     data = pcod_2011,
     mesh = pcod_mesh_2011,
     family = tweedie()
 )
visreg::visreg(fit, xvar = "depth_scaled")

^{Created on 2024-09-03 with reprex v2.1.1}

It should be running the model through predict with newdata and also calculating residuals, but not refitting the model.

visreg has been a big pain to support. You can also make your own visualizations by forming a data frame with all but the variable of interest held at reference values, the variable of interest across a sequence of values, re_form = NA to omit random fields, re_form_iid = NA if needed, and se_fit = TRUE if you want standard errors (will be slower). There's an example in ?predict.sdmTMB if you read down to # Visualizing a marginal effect

[mdmazur]

Thank you, the data I am working with contains fishery data, so not sure I can share a reproducible example here, but here is the code for the sdmTMB model and visreg:

m <- sdmTMB(
  data = res,
  formula = presence ~ s(year,k=4) + poly(log(depth),2) + s(slope,k=5) + s(tpi,k=5) + as.factor(substrate) + as.factor(survey) + s(month,bs='cc',k=12),
  time = "year", mesh = mesh, family = binomial(link = "cloglog"),
  spatial='on',
  spatiotemporal='off',
  anisotropy=TRUE,
  silent=FALSE,
  control=sdmTMBcontrol(newton_loops=1))

visreg::visreg(m,xvar='depth')

Since posting this issue, I found that I don't always get this error. If I close out of my R session and start a new one and run the same code, visreg sometimes works. So maybe this is an issue with my R sessions and not necessarily sdmTMB and visreg.

Thanks for the alternative suggestion on visualizing a marginal effect. I might end up doing that if visreg does not consistently work with my code.

[seananderson]

Closing this since it doesn't seem to be reproducible, but feel free to re-open with a reproducible example.