plot.mgcv.smooth.1d conditional density artefacts about mgcviz HOT 2 CLOSED

I have observed similar problems on other data set. No idea right now.

from mgcviz.

It turns out that in this line

      estXY$fhat <- estXY$fhat / 
                     ( bkde(.datR[ , 1], gridsize = args.dens.opts$ngr[1], 
                            range.x = P$xlim, bandwidth = bw[1])$y  )

the conditional density in the denominator can be almost zero. This in not quite a numerical issue, it's just due to the fact that the kernel density estimator of the marginal density of x goes to zero if no nearby observation is present. Solved by adding a small constant

tmp <- 1e-8 / sqrt(2*pi*var(.datR[ , 1]))

which has the nice property of taking into account the scale of x (via var(datR)). But this solution might not be bullet proof. Similar issue appeared in cdCheck. Both addressed in 4747af1

from mgcviz.