biviso {Iso} | R Documentation |
Bivariate isotonic regression with respect to simple (increasing) linear ordering on both variables.
biviso(y, w = NULL, eps = 1e-04, ncycle = 1000)
y |
The matrix of observations to be isotonized. It must of course have at least two rows and at least two columns. |
w |
A matrix of weights, greater than or equal to zero, of the same
dimension as y . If left NULL then w is
created as a matrix all of whose entries are equal to 1 .
|
eps |
Convergence criterion. The algorithm is deemed to have converged
if each entry of the output matrix, after the completion of the
current iteration, does not differ by more than eps from
the corresponding entry of the matrix after the completion of the
previous iteration.
|
ncycle |
The maximum number of cycles of the iteration procedure. If the
procedure has not converged after ncycle iterations then
an error is given.
|
See the paper by Bril et al., (References) and the references cited therein for details.
A matrix of the same dimensions as y
containing the
corresponding isotonic values. It has an attribute icycle
equal to the number of cycles required to achieve convergence
of the algorithm.
The subroutine comprising Algorithm AS 206 produces an error
code ifault
with values from 1
to 6
The meaning of these codes is as follows:
ncycle
cycles.
w
was negative.
nrow(y)
or ncol(y)
was less than 2.
delta=0.00001
was
replaced by delta
.
delta
.
w
were less than delta
.
If ifault==4
a warning is given. All of the other non-zero
values of ifault
result in an error being given.
Rolf Turner r.turner@auckland.ac.nz http://www.math.unb.ca
Bril, Gordon; Dykstra, Richard; Pillers Carolyn, and Robertson, Tim ; Isotonic regression in two independent variables; Algorithm AS 206; JRSSC (Applied Statistics), vol. 33, no. 3, pp. 352-357, 1984.
x <- 1:20 y <- 1:10 xy <- outer(x,y,function(a,b){a+b+0.5*a*b}) ixy <- biviso(xy) u <- matrix(runif(400),20,20) v <- biviso(u)