Using derivatives with random smooths by individual

[TSeuH]

Hello,

I have been trying to get the first derivatives for each individual fit from the following model:

gam.smooth <- gam(TV ~ groupID +
s(time, by = groupID) +
s(time, animalID, bs = "fs", m = 1),
data = dataTV, method = "REML")

fdSmooth <- derivatives(gam.smooth, n = 1000, term = "s(time)", partial_match = TRUE)

I see a column "fs_var" in fdSmooth, and I was assuming it would contain "animalID", just like by_var contains "groupID". I thought I could get the first derivative for each animalID, but with this code, I only get the first derivative for each groupID, regardless of the animalID.

Would it be possible to have the derivatives by animalID without changing the GAM?

Thank you.

[gavinsimpson]

I will take a closer look later, but initial thoughts are that you need term = "s(time,animalID)" as that's how {mgcv} refers to the smooth and the pattern "s(time)" doesn't match because the closing parenthesis is not after time for the fs smooth

[gavinsimpson]

I'll also add that I doubt the derivative will be that informative if you set m=1. For more reliable estimation of derivatives you need m > d where d is the order of derivative you wish to estimate.

[TSeuH]

Thank you for your answer!
Unfortunately, I get an error message when I try your suggestion:

fdSmooth <- derivatives(gam.smooth, n = 1000, term = "s(time, animalID)", partial_match = TRUE)
Erreur : Failed to match any smooths in model.
Try with 'partial_match = TRUE'?

[TSeuH]

I have just noticed that if I remove the space between time and animalID, I get a different error message:

derivatives(gam.smooth, n = 1000, term = "s(time,animalID)", partial_match = TRUE)
Error in str_detect():
! Can't recycle string (size 2) to match pattern (size 3).
Run rlang::last_trace() to see where the error occurred.

[gavinsimpson]

Can you try that last one without the partial_match = TRUE? You don't need that when you are specifying the full name of the smooth.

[TSeuH]

Thank you for the explanations. Unfortunately, it gives the same error on str_detect().

[gavinsimpson]

I can't reproduce the issue with an example that is similar to the one you describe: see the example I posted in #243

I may have inadvertently fixed this. I'm going to close #243 now. If your problem persists with the CRAN version of gratia, please follow up here or in #243, but I'm going to need a reproducible example I think.