RealizationSink-class: RealizationSink objects

In Bioconductor/DelayedArray: A unified framework for working transparently with on-disk and in-memory array-like datasets

RealizationSink objects

Description

Use a RealizationSink object in combination with write_block() to write blocks of array data to disk.

RealizationSink is a virtual class with various concrete subclasses that support writing data into specific formats.

sinkApply() is a convenience function for walking on a RealizationSink object, typically for the purpose of filling it with blocks of data.

Note that write_block() is typically used inside the callback function passed to sinkApply().

Usage

## Walk on a RealizationSink derivative:
sinkApply(sink, FUN, ..., grid=NULL, verbose=NA)

## Backend-agnostic RealizationSink constructor:
AutoRealizationSink(dim, dimnames=NULL, type="double", as.sparse=FALSE)

## Get/set the "automatic realization backend":
getAutoRealizationBackend()
setAutoRealizationBackend(BACKEND=NULL)
registeredRealizationBackends()

Arguments

sink	A **writable** array-like object, typically a RealizationSink derivative. Some important notes: DelayedArray objects are NEVER writable, even when they don't carry delayed operations (e.g. HDF5Array objects from the HDF5Array package), and even when they don't carry delayed operations **and** have all their data in memory (e.g. RleArray objects). In other words, there are NO exceptions. RealizationSink is a **virtual** class so sink will always be a RealizationSink **derivative**, that is, an object that belongs to a **concrete** subclass of the RealizationSink class (e.g. an HDF5RealizationSink object from the HDF5Array package). RealizationSink derivatives are considered array-like objects i.e. they have dimensions and possibly dimnames. Although write_block() and sinkApply() will typically be used on a RealizationSink derivative, they can also be used on an ordinary array or other writable in-memory array-like objects like dgCMatrix objects from the Matrix package.
FUN	The callback function to apply to each **viewport** of the grid used to walk on sink. sinkApply() will perform sink <- FUN(sink, viewport, ...) on each viewport, so FUN must take at least two arguments, typically sink and viewport (but the exact names can differ). The function is expected to return its 1st argument (sink) possibly modified (e.g. when FUN contains a call to write_block(), which is typically the case).
...	Additional arguments passed to FUN.
grid	The grid used for the walk, that is, an ArrayGrid object that defines the viewports to walk on. It must be compatible with the geometry of sink. If not specified, an automatic grid is created by calling defaultSinkAutoGrid(sink), and used. See ?defaultSinkAutoGrid for more information.
verbose	Whether block processing progress should be displayed or not. If set to NA (the default), verbosity is controlled by DelayedArray:::get_verbose_block_processing(). Setting verbose to TRUE or FALSE overrides this.
dim	The dimensions (specified as an integer vector) of the RealizationSink derivative to create.
dimnames	The dimnames (specified as a list of character vectors or NULLs) of the RealizationSink derivative to create.
type	The type of the data that will be written to the RealizationSink derivative to create.
as.sparse	Whether the data should be written as sparse or not to the RealizationSink derivative to create. Not all realization backends support this.
BACKEND	NULL (the default), or a single string specifying the name of a realization backend e.g. "HDF5Array" or "RleArray" etc...

Details

*** The RealizationSink API ***

The DelayedArray package provides a simple API for writing blocks of array data to disk (or to memory): the "RealizationSink API". This API allows the developper to write code that is agnostic about the particular on-disk (or in-memory) format being used to store the data.

Here is how to use it:

1. Create a realization sink.

2. Write blocks of array data to the realization sink with one or several calls to write_block().

3. Close the realization sink with close().

4. Coerce the realization sink to DelayedArray.

A realization sink is formally represented by a RealizationSink derivative. Note that RealizationSink is a virtual class with various concrete subclasses like HDF5RealizationSink from the HDF5Array package, or RleRealizationSink. Each subclass implements the "RealizationSink API" for a specific realization backend.

To create a realization sink, use the specific constructor function. This function should be named as the class itself e.g. HDF5RealizationSink().

To create a realization sink in a backend-agnostic way, use AutoRealizationSink(). It will create a RealizationSink derivative for the current automatic realization backend (see below).

Once writing to the realization sink is completed, the RealizationSink derivative must be closed (with close(sink)), then coerced to DelayedArray (with as(sink, "DelayedArray"). What specific DelayedArray derivative this coercion will return depends on the specific class of the RealizationSink derivative. For example, if sink is an HDF5RealizationSink object from the HDF5Array package, then as(sink, "DelayedArray") will return an HDF5Array instance (the HDF5Array class is a DelayedArray subclass).

*** The automatic realization backend ***

The automatic realization backend is a user-controlled global setting that indicates what specific RealizationSink derivative AutoRealizationSink() should return. In the context of block processing of a DelayedArray object, this controls where/how realization happens e.g. as an ordinary array if not set (i.e. set to NULL), or as an HDF5Array object if set to "HDF5Array", or as an RleArray object if set to "RleArray", etc...

Use getAutoRealizationBackend() or setAutoRealizationBackend() to get or set the automatic realization backend.

Use registeredRealizationBackends() to get the list of realization backends that are currently registered.

*** Cross realization backend compatibility ***

Two important things to keep in mind for developers aiming at writing code that is compatible across realization backends:

• Realization backends don't necessarily support concurrent writing.

  More precisely: Even though it is safe to assume that any DelayedArray object will support concurrent read_block() calls, it is not so safe to assume that any RealizationSink derivative will support concurrent calls to write_block(). For example, at the moment, HDF5RealizationSink objects do not support concurrent writing.

  This means that in order to remain compatible across realization backends, code that contains calls to write_block() should NOT be parallelized.

• Some realization backends are "linear write only", that is, they don't support random write access, only linear write access.

  Such backends will provide a relization sink where the blocks of data must be written in linear order (i.e. by ascending rank). Furthermore, the geometry of the blocks must also be compatible with linear write access, that is, they must have a "first-dim-grows-first" shape. Concretely this means that the grid used to walk on the relization sink must be created with something like:

      colAutoGrid(sink)

  for a two-dimensional sink, or with something like:

      defaultSinkAutoGrid(sink)

  for a sink with an arbitrary number of dimensions.

  See ?defaultSinkAutoGrid for more information.

  For obvious reasons, "linear write only" realization backends do not support concurrent writing.

Value

For sinkApply(), its 1st argument (sink) possibly modified (e.g. when callback function FUN contains a call to write_block(), which is typically the case).

For AutoRealizationSink(), a RealizationSink derivative with the class associated with the current automatic realization backend.

For getAutoRealizationBackend, NULL (no backend set yet) or a single string specifying the name of the automatic realization backend currently in use.

For registeredRealizationBackends, a data frame with 1 row per registered realization backend.

See Also

• read_block and write_block in the S4Arrays package.

• ArrayGrid in the S4Arrays package for the formal representation of grids and viewports.

• defaultSinkAutoGrid to create an automatic grid on a RealizationSink derivative.

• blockApply and family for convenient block processing of an array-like object.

• HDF5RealizationSink objects in the HDF5Array package.

• HDF5-dump-management in the HDF5Array package to control the location and physical properties of automatically created HDF5 datasets.

• RleArray objects.

• DelayedArray objects.

• array objects in base R.

Examples

## ---------------------------------------------------------------------
## USING THE "RealizationSink API": EXAMPLE 1
## ---------------------------------------------------------------------

## -- STEP 1 --
## Create a realization sink. Note that instead of creating a
## realization sink by calling a backend-specific sink constructor
## (e.g. HDF5Array::HDF5RealizationSink), we set the "automatic
## realization backend" to "HDF5Array" and use backend-agnostic
## constructor AutoRealizationSink():
setAutoRealizationBackend("HDF5Array")
sink <- AutoRealizationSink(c(35L, 50L, 8L))
dim(sink)

## -- STEP 2 --
## Define the grid of viewports to walk on. Here we define a grid made
## of very small viewports on 'sink'. Note that, for real-world use cases,
## block processing will typically use grids made of much bigger
## viewports, usually obtained with defaultSinkAutoGrid().
## Also please note that this grid would not be compatible with "linear
## write only" realization backends. See "Cross realization backend
## compatibility" above in this man page for more information.
sink_grid <- RegularArrayGrid(dim(sink), spacings=c(20, 20, 4))

## -- STEP 3 --
## Walk on the grid, and, for each viewport, write random data to it.
for (bid in seq_along(sink_grid)) {
    viewport <- sink_grid[[bid]]
    block <- array(runif(length(viewport)), dim=dim(viewport))
    sink <- write_block(sink, viewport, block)
}

## -- An alternative to STEP 3 --
FUN <- function(sink, viewport) {
    block <- array(runif(length(viewport)), dim=dim(viewport))
    write_block(sink, viewport, block)
}
sink <- sinkApply(sink, FUN, grid=sink_grid, verbose=TRUE)

## -- STEP 4 --
## Close the sink and turn it into a DelayedArray object:
close(sink)
A <- as(sink, "DelayedArray")
A

setAutoRealizationBackend()  # restore default (NULL)

## ---------------------------------------------------------------------
## USING THE "RealizationSink API": EXAMPLE 2
## ---------------------------------------------------------------------

## Say we have a 3D array and want to collapse its 3rd dimension by
## summing the array elements that are stacked vertically, that is, we
## want to compute the matrix M obtained by doing sum(A[i, j, ]) for all
## valid i and j. This is very easy to do with an ordinary array:
collapse_3rd_dim <- function(a) apply(a, MARGIN=1:2, sum)

## or, in a slightly more efficient way:
collapse_3rd_dim <- function(a) {
    m <- matrix(0, nrow=nrow(a), ncol=ncol(a))
    for (z in seq_len(dim(a)[[3]]))
        m <- m + a[ , , z]
    m
}

## With a toy 3D array:
a <- array(runif(8000), dim=c(25, 40, 8))
dim(collapse_3rd_dim(a))
stopifnot(identical(sum(a), sum(collapse_3rd_dim(a))))  # sanity check

## Now say that A is so big that even M wouldn't fit in memory. This is
## a situation where we'd want to compute M block by block:

## -- STEP 1 --
## Create the 2D realization sink:
setAutoRealizationBackend("HDF5Array")
sink <- AutoRealizationSink(dim(a)[1:2])
dim(sink)

## -- STEP 2 --
## Define two grids: one for 'sink' and one for 'a'. Since we're going
## to walk on the two grids simultaneously, read a block from 'a' and
## write it to 'sink', we need to make sure that we define grids that
## are "aligned". More precisely, the two grids must have the same number
## of viewports, and the viewports in one must correspond to the viewports
## in the other one:
sink_grid <- colAutoGrid(sink, ncol=10)
a_spacings <- c(dim(sink_grid[[1L]]), dim(a)[[3]])
a_grid <- RegularArrayGrid(dim(a), spacings=a_spacings)
dims(sink_grid)  # dimensions of the individual viewports
dims(a_grid)     # dimensions of the individual viewports

## Let's check that our two grids are actually "aligned":
stopifnot(identical(length(sink_grid), length(a_grid)))
stopifnot(identical(dims(sink_grid), dims(a_grid)[ , 1:2, drop=FALSE]))

## -- STEP 3 --
## Walk on the two grids simultaneously:
for (bid in seq_along(sink_grid)) {
    ## Read block from 'a'.
    a_viewport <- a_grid[[bid]]
    block <- read_block(a, a_viewport)
    ## Collapse it.
    block <- collapse_3rd_dim(block)
    ## Write the collapsed block to 'sink'.
    sink_viewport <- sink_grid[[bid]]
    sink <- write_block(sink, sink_viewport, block)
}

## -- An alternative to STEP 3 --
FUN <- function(sink, sink_viewport) {
    ## Read block from 'a'.
    bid <- currentBlockId()
    a_viewport <- a_grid[[bid]]
    block <- read_block(a, a_viewport)
    ## Collapse it.
    block <- collapse_3rd_dim(block)
    ## Write the collapsed block to 'sink'.
    write_block(sink, sink_viewport, block)
}
sink <- sinkApply(sink, FUN, grid=sink_grid, verbose=TRUE)

## -- STEP 4 --
## Close the sink and turn it into a DelayedArray object:
close(sink)
M <- as(sink, "DelayedArray")
M

## Sanity check:
stopifnot(identical(collapse_3rd_dim(a), as.array(M)))

setAutoRealizationBackend()  # restore default (NULL)

## ---------------------------------------------------------------------
## USING THE "RealizationSink API": AN ADVANCED EXAMPLE
## ---------------------------------------------------------------------

## Say we have 2 matrices with the same number of columns. Each column
## represents a biological sample:
library(HDF5Array)
R <- as(matrix(runif(75000), ncol=1000), "HDF5Array")   # 75 rows
G <- as(matrix(runif(250000), ncol=1000), "HDF5Array")  # 250 rows

## Say we want to compute the matrix U obtained by applying the same
## binary functions FUN() to all samples i.e. U is defined as:
##
##   U[ , j] <- FUN(R[ , j], G[ , j]) for 1 <= j <= 1000
##
## Note that FUN() should return a vector of constant length, say 200,
## so U will be a 200x1000 matrix. A naive implementation would be:
##
##   pFUN <- function(r, g) {
##       stopifnot(ncol(r) == ncol(g))  # sanity check
##       sapply(seq_len(ncol(r)), function(j) FUN(r[ , j], g[ , j]))
##   }
##
## But because U is going to be too big to fit in memory, we can't
## just do pFUN(R, G). So we want to compute U block by block and
## write the blocks to disk as we go. The blocks will be made of full
## columns. Also since we need to walk on 2 matrices at the same time
## (R and G), we can't use blockApply() or blockReduce() so we'll use
## a "for" loop.

## Before we get to the "for" loop, we need 4 things:

## 1. Two grids of blocks, one on R and one on G. The blocks in the
##    two grids must contain the same number of columns. We arbitrarily
##    choose to use blocks of 150 columns:
R_grid <- colAutoGrid(R, ncol=150)
G_grid <- colAutoGrid(G, ncol=150)

## 2. The function pFUN(). It will take 2 blocks as input, 1 from R
##    and 1 from G, apply FUN() to all the samples in the blocks,
##    and return a matrix with one columns per sample:
pFUN <- function(r, g) {
    stopifnot(ncol(r) == ncol(g))  # sanity check
    ## Return a matrix with 200 rows with random values. Completely
    ## artificial sorry. A realistic example would actually need to
    ## apply the same binary function to r[ ,j] and g[ , j] for
    ## 1 <= j <= ncol(r).
    matrix(runif(200 * ncol(r)), nrow=200)
}

## 3. A RealizationSink derivative where to write the matrices returned
##    by pFUN() as we go:
setAutoRealizationBackend("HDF5Array")
U_sink <- AutoRealizationSink(c(200L, 1000L))

## 4. Finally, we create a grid on U_sink with viewports that contain
##    the same number of columns as the corresponding blocks in R and G:
U_grid <- colAutoGrid(U_sink, ncol=150)

## Note that the three grids should have the same number of viewports:
stopifnot(length(U_grid) == length(R_grid))
stopifnot(length(U_grid) == length(G_grid))

## 5. Now we can proceed. We use a "for" loop to walk on R and G
##    simultaneously, block by block, apply pFUN(), and write the
##    output of pFUN() to U_sink:
for (bid in seq_along(U_grid)) {
    R_block <- read_block(R, R_grid[[bid]])
    G_block <- read_block(G, G_grid[[bid]])
    U_block <- pFUN(R_block, G_block)
    U_sink <- write_block(U_sink, U_grid[[bid]], U_block)
}

## An alternative to the "for" loop is to use sinkApply():
FUN <- function(U_sink, U_viewport) {
    bid <- currentBlockId()
    R_block <- read_block(R, R_grid[[bid]])
    G_block <- read_block(G, G_grid[[bid]])
    U_block <- pFUN(R_block, G_block)
    write_block(U_sink, U_viewport, U_block)
}
U_sink <- sinkApply(U_sink, FUN, grid=U_grid, verbose=TRUE)

close(U_sink)
U <- as(U_sink, "DelayedArray")
U

setAutoRealizationBackend()  # restore default (NULL)

## ---------------------------------------------------------------------
## VERY BASIC (BUT ALSO VERY ARTIFICIAL) USAGE OF THE
## read_block()/write_block() COMBO
## ---------------------------------------------------------------------

###### On an ordinary matrix ######
m1 <- matrix(1:30, ncol=5)

## Define a viewport on 'm1':
block1_dim <- c(4, 3)
viewport1 <- ArrayViewport(dim(m1), IRanges(c(3, 2), width=block1_dim))

## Read/tranform/write:
block1 <- read_block(m1, viewport1)
write_block(m1, viewport1, block1 + 1000L)

## Define another viewport on 'm1':
viewport1b <- ArrayViewport(dim(m1), IRanges(c(1, 3), width=block1_dim))

## Read/tranform/write:
write_block(m1, viewport1b, block1 + 1000L)

## No-op:
m <- write_block(m1, viewport1, read_block(m1, viewport1))
stopifnot(identical(m1, m))

########## On a 3D array ##########
a3 <- array(1:60, 5:3)

## Define a viewport on 'a3':
block3_dim <- c(2, 4, 1)
viewport3 <- ArrayViewport(dim(a3), IRanges(c(1, 1, 3), width=block3_dim))

## Read/tranform/write:
block3 <- read_block(a3, viewport3)
write_block(a3, viewport3, block3 + 1000L)

## Define another viewport on 'a3':
viewport3b <- ArrayViewport(dim(a3), IRanges(c(3, 1, 3), width=block3_dim))

## Read/tranform/write:
write_block(a3, viewport3b, block3 + 1000L)

## No-op:
a <- write_block(a3, viewport3, read_block(a3, viewport3))
stopifnot(identical(a3, a))

## ---------------------------------------------------------------------
## LESS BASIC (BUT STILL VERY ARTIFICIAL) USAGE OF THE
## read_block()/write_block() COMBO
## ---------------------------------------------------------------------

grid1 <- RegularArrayGrid(dim(m1), spacings=c(3L, 2L))
grid1
length(grid1)  # number of blocks defined by the grid
read_block(m1, grid1[[3L]])  # read 3rd block
read_block(m1, grid1[[1L, 3L]])

## Walk on the grid, colum by column:
m1a <- m1
for (bid in seq_along(grid1)) {
    viewport <- grid1[[bid]]
    block <- read_block(m1a, viewport)
    block <- bid * 1000L + block
    m1a <- write_block(m1a, viewport, block)
}
m1a

## Walk on the grid, row by row:
m1b <- m1
for (i in seq_len(dim(grid1)[[1]])) {
  for (j in seq_len(dim(grid1)[[2]])) {
    viewport <- grid1[[i, j]]
    block <- read_block(m1b, viewport)
    block <- (i * 10L + j) * 1000L + block
    m1b <- write_block(m1b, viewport, block)
  }
}
m1b

## ---------------------------------------------------------------------
## registeredRealizationBackends() AND FAMILY
## ---------------------------------------------------------------------

getAutoRealizationBackend()  # no backend set yet

registeredRealizationBackends()
setAutoRealizationBackend("HDF5Array")
getAutoRealizationBackend()  # backend is set to "HDF5Array"
registeredRealizationBackends()

getHDF5DumpChunkLength()
setHDF5DumpChunkLength(500L)
getHDF5DumpChunkShape()

sink <- AutoRealizationSink(c(120L, 50L))
class(sink)  # HDF5-specific realization sink
dim(sink)
chunkdim(sink)

grid <- defaultSinkAutoGrid(sink, block.length=600)
for (bid in seq_along(grid)) {
    viewport <- grid[[bid]]
    block <- 101 * bid + runif(length(viewport))
    dim(block) <- dim(viewport)
    sink <- write_block(sink, viewport, block)
}

close(sink)
A <- as(sink, "DelayedArray")
A

setAutoRealizationBackend()  # restore default (NULL)

Bioconductor/DelayedArray documentation built on Nov. 18, 2024, 3:06 a.m.