Binds arrays corner-to-corner
Array generalization of blockdiag()
adiag(... , pad=as.integer(0), do.dimnames=TRUE)
... |
Arrays to be binded together |
pad |
Value to pad array with; note default keeps integer status of arrays |
do.dimnames |
Boolean, with default |
Binds any number of arrays together, corner-to-corner. Because the
function is associative provided pad is of length 1, this page
discusses the two array case.
Suppose x <- adiag(a,b) and dim(a)=c(a_1,...,a_d),
dim(b)=c(b_1,...,b_d). Then we have
all(dim(x)==dim(a)+dim(b)); and x[1:a_1,...,1:a_d]==a and
x[(a_1+1):(a_1+b_1),...,(a_d+1):(a_d+b_d)]==b.
Dimnames are preserved, if both arrays have non-null dimnames, and
do.dimnames is TRUE.
Argument pad is usually a length-one vector, but any vector is
acceptable; standard recycling is used. Be aware that the output array
(of dimension dim(a)+dim(b)) is filled with (copies of)
pad before a and b are copied. This can be
confusing.
Returns an array of dimensions dim(a)+dim(b) as described above.
In adiag(a,b), if a is a length-one vector, it is coerced
to an array of dimensions rep(1,length(dim(b))); likewise
b. If both a and b are length-one vectors, return
diag(c(a,b)).
If a and b are arrays, function adiag() requires
length(dim(a))==length(dim(b)) (the function does not guess which
dimensions have been dropped; see examples section). In particular,
note that vectors are not coerced except if of length one.
adiag() is used when padding magic hypercubes in the context
of evaluating subarray sums.
Peter Wolf with some additions by Robin Hankin
a <- array( 1,c(2,2))
b <- array(-1,c(2,2))
adiag(a,b)
## dropped dimensions can count:
b2 <- b1 <- b
dim(a) <- c(2,1,2)
dim(b1) <- c(2,2,1)
dim(b2) <- c(1,2,2)
dim(adiag(a,b1))
dim(adiag(a,b2))
## dimnames are preserved if not null:
a <- matrix(1,2,2,dimnames=list(col=c("red","blue"),size=c("big","small")))
b <- 8
dim(b) <- c(1,1)
dimnames(b) <- list(col=c("green"),size=c("tiny"))
adiag(a,b) #dimnames preserved
adiag(a,8) #dimnames lost because second argument has none.
## non scalar values for pad can be confusing:
q <- matrix(0,3,3)
adiag(q,q,pad=1:4)
## following example should make the pattern clear:
adiag(q,q,pad=1:36)
# Now, a use for arrays with dimensions of zero extent:
z <- array(dim=c(0,3))
colnames(z) <- c("foo","bar","baz")
adiag(a,z) # Observe how this has
# added no (ie zero) rows to "a" but
# three extra columns filled with the pad value
adiag(a,t(z))
adiag(z,t(z)) # just the pad valuePlease choose more modern alternatives, such as Google Chrome or Mozilla Firefox.