simplify {DelayedArray} | R Documentation |
Simplify a tree of delayed operations
Description
NOTE: The tools documented in this man page are primarily intended for developers or advanced users curious about the internals of the DelayedArray package. End users typically don't need them for their regular use of DelayedArray objects.
In a DelayedArray object, the delayed operations are stored as a
tree of DelayedOp objects. See ?DelayedOp
for more
information about this tree.
simplify
can be used to simplify the tree of delayed operations
in a DelayedArray object.
isPristine
can be used to know whether a DelayedArray
object is pristine or not. A DelayedArray object is
considered pristine when it carries no delayed operation.
Note that an object that carries delayed operations that do nothing
(e.g. A + 0
) is not considered pristine.
contentIsPristine
can be used to know whether the delayed
operations in a DelayedArray object touch its array
elements or not.
netSubsetAndAperm
returns an object that represents the net
subsetting and net dimension rearrangement of all the delayed
operations in a DelayedArray object.
Usage
simplify(x, incremental=FALSE)
isPristine(x, ignore.dimnames=FALSE)
contentIsPristine(x)
netSubsetAndAperm(x, as.DelayedOp=FALSE)
Arguments
x |
Typically a DelayedArray object but can also be a DelayedOp
object (except for |
incremental |
For internal use. |
ignore.dimnames |
|
as.DelayedOp |
|
Details
netSubsetAndAperm
is only supported on a DelayedArray
object x
with a single seed i.e. if nseed(x) == 1
.
The mapping between the array elements of x
and the array elements
of its seed is affected by the following delayed operations carried by
x
: [
, drop()
, and aperm()
.
x
can carry any number of each of these operations in any order but
their net result can always be described by a net subsetting
followed by a net dimension rearrangement.
netSubsetAndAperm(x)
returns an object that represents the
net subsetting and net dimension rearrangement.
The as.DelayedOp
argument controls in what form this object should
be returned:
If
as.DelayedOp
isFALSE
(the default), the returned object is a list of subscripts that describes the net subsetting. The list contains one subscript per dimension in the seed. Each subscript can be either a vector of positive integers or aNULL
. ANULL
indicates a missing subscript. In addition, ifx
carries delayed operations that rearrange its dimensions (i.e. operations that drop and/or permute some of the original dimensions), the net dimension rearrangement is described in adimmap
attribute added to the list. This attribute is an integer vector parallel todim(x)
that reports how the dimensions ofx
are mapped to the dimensions of its seed.If
as.DelayedOp
isTRUE
, the returned object is a linear tree with 2 DelayedOp nodes and a leaf node. The leaf node is the seed ofx
. Walking the tree from the seed, the 2 DelayedOp nodes are of type DelayedSubset and DelayedAperm, in that order (this reflects the order in which the operations apply). More precisely, the returned object is a DelayedAperm object with one child (the DelayedSubset object), and one grandchid (the seed ofx
). The DelayedSubset and DelayedAperm nodes represent the net subsetting and net dimension rearrangement, respectively. Either or both of them can be a no-op.
Note that the returned object describes how the array elements of x
map to their corresponding array element in seed(x)
.
Value
The simplified object for simplify
.
TRUE
or FALSE
for contentIsPristine
.
An ordinary list (possibly with the dimmap
attribute on it) for
netSubsetAndAperm
. Unless as.DelayedOp
is set to TRUE
,
in which case a DelayedAperm object is returned (see Details
section above for more information).
See Also
-
showtree
to visualize and access the leaves of a tree of delayed operations carried by a DelayedArray object. -
DelayedOp objects.
-
DelayedArray objects.
Examples
## ---------------------------------------------------------------------
## Simplification of the tree of delayed operations
## ---------------------------------------------------------------------
m1 <- matrix(runif(150), nrow=15, ncol=10)
M1 <- DelayedArray(m1)
showtree(M1)
## By default, the tree of delayed operations carried by a DelayedArray
## object gets simplified each time a delayed operation is added to it.
## This can be disabled via a global option:
options(DelayedArray.simplify=FALSE)
M2 <- log(t(M1[5:1, c(TRUE, FALSE)] + 10))[-1, ]
showtree(M2) # linear tree
## Note that as part of the simplification process, some operations
## can be reordered:
options(DelayedArray.simplify=TRUE)
M2 <- log(t(M1[5:1, c(TRUE, FALSE)] + 10))[-1, ]
showtree(M2) # linear tree
options(DelayedArray.simplify=FALSE)
dimnames(M1) <- list(letters[1:15], LETTERS[1:10])
showtree(M1) # linear tree
m2 <- matrix(1:20, nrow=10)
Y <- cbind(t(M1[ , 10:1]), DelayedArray(m2), M1[6:15, "A", drop=FALSE])
showtree(Y) # non-linear tree
Z <- t(Y[10:1, ])[1:15, ] + 0.4 * M1
showtree(Z) # non-linear tree
Z@seed@seeds
Z@seed@seeds[[2]]@seed # reaching to M1
Z@seed@seeds[[1]]@seed@seed@seed@seed@seed # reaching to Y
## ---------------------------------------------------------------------
## isPristine()
## ---------------------------------------------------------------------
m <- matrix(1:20, ncol=4, dimnames=list(letters[1:5], NULL))
M <- DelayedArray(m)
isPristine(M) # TRUE
isPristine(log(M)) # FALSE
isPristine(M + 0) # FALSE
isPristine(t(M)) # FALSE
isPristine(t(t(M))) # TRUE
isPristine(cbind(M, M)) # FALSE
isPristine(cbind(M)) # TRUE
dimnames(M) <- NULL
isPristine(M) # FALSE
isPristine(M, ignore.dimnames=TRUE) # TRUE
isPristine(t(t(M)), ignore.dimnames=TRUE) # TRUE
isPristine(cbind(M, M), ignore.dimnames=TRUE) # FALSE
## ---------------------------------------------------------------------
## contentIsPristine()
## ---------------------------------------------------------------------
a <- array(1:40, c(4, 5, 2))
A <- DelayedArray(a)
stopifnot(contentIsPristine(A))
stopifnot(contentIsPristine(A[1, , ]))
stopifnot(contentIsPristine(t(A[1, , ])))
stopifnot(contentIsPristine(cbind(A[1, , ], A[2, , ])))
dimnames(A) <- list(LETTERS[1:4], letters[1:5], NULL)
stopifnot(contentIsPristine(A))
contentIsPristine(log(A)) # FALSE
contentIsPristine(A - 11:14) # FALSE
contentIsPristine(A * A) # FALSE
## ---------------------------------------------------------------------
## netSubsetAndAperm()
## ---------------------------------------------------------------------
a <- array(1:40, c(4, 5, 2))
M <- aperm(DelayedArray(a)[ , -1, ] / 100)[ , , 3] + 99:98
M
showtree(M)
netSubsetAndAperm(M) # 1st dimension was dropped, 2nd and 3rd
# dimension were permuted (transposition)
op2 <- netSubsetAndAperm(M, as.DelayedOp=TRUE)
op2 # 2 nested delayed operations
op1 <- op2@seed
class(op1) # DelayedSubset
class(op2) # DelayedAperm
op1@index
op2@perm
DelayedArray(op2) # same as M from a [, drop(), and aperm() point of
# view but the individual array elements are now
# reset to their original values i.e. to the values
# they have in the seed
stopifnot(contentIsPristine(DelayedArray(op2)))
## A simple function that returns TRUE if a DelayedArray object carries
## no "net subsetting" and no "net dimension rearrangement":
is_aligned_with_seed <- function(x)
{
if (nseed(x) != 1L)
return(FALSE)
op2 <- netSubsetAndAperm(x, as.DelayedOp=TRUE)
op1 <- op2@seed
is_noop(op1) && is_noop(op2)
}
M <- DelayedArray(a[ , , 1])
is_aligned_with_seed(log(M + 11:14) > 3) # TRUE
is_aligned_with_seed(M[4:1, ]) # FALSE
is_aligned_with_seed(M[4:1, ][4:1, ]) # TRUE
is_aligned_with_seed(t(M)) # FALSE
is_aligned_with_seed(t(t(M))) # TRUE
is_aligned_with_seed(t(0.5 * t(M[4:1, ])[ , 4:1])) # TRUE
options(DelayedArray.simplify=TRUE)