derivatives can't handle by variable smooths

[bart-larsen]

I really like the derivatives function, but I have been running into this issue.
The issue occurs whether I use ordered or unordered factors in the by-factor smooths. I have been mostly testing this with gamm, but it seems to not work for gam either.
For example:

modobj <- gamm(y ~ s(age,by = sex),random=list(ID=~1)
df <- derivatives(modobj,'age)

This returns the derivative for the first level of the by factor smooth, but returns all zeros for the derivative and standard error for the next level.

If I run this in the ordered factor case to test for interactions,
modobj <- gamm(y ~ s(age) +oSex + s(age,by = oSex),random=list(ID=~1)
df <- derivatives(modobj,'age)

It returns the derivative for the reference smooth (s(age)), but all zeros for the derivative and standard error of the by factor smooth.

[gavinsimpson]

Here's a reproducible example using gam()

set.seed(1)
dat <- gamSim(4, n = 400, verbose = FALSE)
m <- gam(y ~ fac + s(x2, by = fac), data = dat)
d <- derivatives(m)
draw(d) # clearly shows the problem

and another one for testing

m2 <- gam(y ~ x1 + s(x2) + fac + s(x0, by = fac), data = dat)

and another one for testing, this time ordered factor

dat <- transform(dat, ofac = ordered(fac))
m3 <- gam(y ~ x1 + s(x2) + ofac + s(x0) + s(x0, by = ofac), data = dat)

[gavinsimpson]

An immediate issue is that derivative_data() is only returning data for the first level of the factor. This needs fixing.

[gavinsimpson]

This is now working for both ordered and ordinary factor by smooths for gam() and gamm() models.

Thanks for the report.