This discrete kernel smoother uses the center cell, immediate
neighbors and second neighbors to smooth the counts. The counts for
each resulting cell is a linear combination of previous cell counts
and weights. The weights are
ll{
1 center cell, weight = wts[1]
6 immediate neighbors weight = wts[2]
12 second neighbors weight =wts[3]
}
If a cell, its immediate and second neighbors all have a value of
max(cnt), the new maximum count would be
max(cnt)*sum(wts). It is possible for the counts to overflow. The domain for cells with positive counts increases. The hexbin
slots xbins, xbnds, ybnds, and dimen all
reflect this increase.
Note that usually dimen[2] = xbins+1.
The intent was to provide a fast, iterated, immediate neighbor smoother.
However, the current hexbin plotting routines only support shifting
even numbered rows to the right. Future work can
(1) add a shift indicator to hexbin objects that indicates left or
right shifting.
(2) generalize plot.hexbin() and hexagons()
(3) provide an iterated kernel.
With wts[3]=0, the smoother only uses the immediate neighbors.
With a shift indicator the domain could increase by 2 rows (one bottom
and on top) and 2 columns (one left and one right). However the current
implementation increases the domain by 4 rows and 4 columns, thus
reducing plotting resolution.