AutoGrid: Generate automatic grids to use for block processing of...

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

Description

defaultAutoGrid() is the default automatic grid maker. It creates a grid that is suitable for block processing of the array-like object passed to it.

rowAutoGrid() and colAutoGrid() are additional automatic grid makers, specific to the 2-dimensional case. They can be used to define blocks of full rows or full columns.

Usage

defaultAutoGrid(x, block.length=NULL, chunk.grid=NULL, block.shape=NULL)

rowAutoGrid(x, nrow=NULL, block.length=NULL)
colAutoGrid(x, ncol=NULL, block.length=NULL)

## Replace default automatic grid maker with user-defined one:
getAutoGridMaker()
setAutoGridMaker(GRIDMAKER="defaultAutoGrid")

Arguments

x	An array-like or matrix-like object for defaultAutoGrid. A matrix-like object for rowAutoGrid and colAutoGrid.
block.length	The length of the blocks i.e. the number of array elements per block. By default the automatic block length (returned by getAutoBlockLength(type(x))) is used. Depending on how much memory is available on your machine, you might want to increase (or decrease) the automatic block length by adjusting the automatic block size with setAutoBlockSize().
chunk.grid	The grid of physical chunks. By default chunkGrid(x) is used.
block.shape	A string specifying the shape of the blocks. See makeCappedVolumeBox for a description of the supported shapes. By default getAutoBlockShape() is used.
nrow	The number of rows of the blocks. The bottommost blocks might have less. See examples below.
ncol	The number of columns of the blocks. The rightmost blocks might have less. See examples below.
GRIDMAKER	The function to use as automatic grid maker, that is, the function that will be used by blockApply() and blockReduce() to make a grid when no grid is supplied via their grid argument. The function will be called on array-like object x and must return an ArrayGrid object, say grid, that is compatible with x i.e. such that refdim(grid) is identical to dim(x). GRIDMAKER can be specified as a function or as a single string naming a function. It can be a user-defined function or a pre-defined grid maker like defaultAutoGrid, rowAutoGrid, or colAutoGrid. The automatic grid maker is set to defaultAutoGrid at package startup and can be reset anytime to this value by calling setAutoGridMaker() with no argument.

Details

By default, primary block processing functions blockApply() and blockReduce() use the grid returned by defaultAutoGrid(x) to process array-like object x block by block. This can be changed with setAutoGridMaker().

Value

defaultAutoGrid: An ArrayGrid object on reference array x. The grid elements define the blocks that will be used to process x by block. The grid is optimal in the sense that:

1. It's compatible with the grid of physical chunks a.k.a. chunk grid. This means that, when the chunk grid is known (i.e. when chunkGrid(x) is not NULL or chunk.grid is supplied), every block in the grid contains one or more full chunks. In other words, chunks never cross block boundaries.

2. Its resolution is such that the blocks have a length that is as close as possibe to (but does not exceed) block.length. An exception is made when some chunks already have a length that is >= block.length, in which case the returned grid is the same as the chunk grid.

Note that the returned grid is regular (i.e. is a RegularArrayGrid object) unless the chunk grid is not regular (i.e. is an ArbitraryArrayGrid object).

rowAutoGrid: A RegularArrayGrid object on reference array x where the grid elements define blocks made of full rows of x.

colAutoGrid: A RegularArrayGrid object on reference array x where the grid elements define blocks made of full columns of x.

See Also

• setAutoBlockSize and setAutoBlockShape to control the geometry of automatic blocks.

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

• ArrayGrid for the formal representation of grids and viewports.

• The makeCappedVolumeBox utility to make capped volume boxes.

• chunkGrid.

• read_block and write_block.

Examples

## ---------------------------------------------------------------------
## A VERSION OF sum() THAT USES BLOCK PROCESSING
## ---------------------------------------------------------------------

block_sum <- function(a, grid) {
  sums <- lapply(grid, function(viewport) sum(read_block(a, viewport)))
  sum(unlist(sums))
}

## On an ordinary matrix:
m <- matrix(runif(600), ncol=12)
m_grid <- defaultAutoGrid(m, block.length=120)
sum1 <- block_sum(m, m_grid)
sum1

## On a DelayedArray object:
library(HDF5Array)
M <- as(m, "HDF5Array")
sum2 <- block_sum(M, m_grid)
sum2

sum3 <- block_sum(M, colAutoGrid(M, block.length=120))
sum3

sum4 <- block_sum(M, rowAutoGrid(M, block.length=80))
sum4

## Sanity checks:
sum0 <- sum(m)
stopifnot(identical(sum1, sum0))
stopifnot(identical(sum2, sum0))
stopifnot(identical(sum3, sum0))
stopifnot(identical(sum4, sum0))

## ---------------------------------------------------------------------
## defaultAutoGrid()
## ---------------------------------------------------------------------
grid <- defaultAutoGrid(m, block.length=120)
grid
as.list(grid)  # turn the grid into a list of ArrayViewport objects
table(lengths(grid))
stopifnot(maxlength(grid) <= 120)

grid <- defaultAutoGrid(m, block.length=120,
                           block.shape="first-dim-grows-first")
grid
table(lengths(grid))
stopifnot(maxlength(grid) <= 120)

grid <- defaultAutoGrid(m, block.length=120,
                           block.shape="last-dim-grows-first")
grid
table(lengths(grid))
stopifnot(maxlength(grid) <= 120)

defaultAutoGrid(m, block.length=100)
defaultAutoGrid(m, block.length=75)
defaultAutoGrid(m, block.length=25)
defaultAutoGrid(m, block.length=20)
defaultAutoGrid(m, block.length=10)

## ---------------------------------------------------------------------
## rowAutoGrid() AND colAutoGrid()
## ---------------------------------------------------------------------
rowAutoGrid(m, nrow=10)  # 5 blocks of 10 rows each
rowAutoGrid(m, nrow=15)  # 3 blocks of 15 rows each plus 1 block of 5 rows
colAutoGrid(m, ncol=5)   # 2 blocks of 5 cols each plus 1 block of 2 cols

## See '?write_block' for an advanced example of user-implemented
## block processing using colAutoGrid() and a realization sink.

## ---------------------------------------------------------------------
## REPLACE DEFAULT AUTOMATIC GRID MAKER WITH USER-DEFINED ONE
## ---------------------------------------------------------------------
getAutoGridMaker()
setAutoGridMaker(function(x) colAutoGrid(x, ncol=5))
getAutoGridMaker()

blockApply(m, function(block) currentViewport())

## Reset automatic grid maker to factory settings:
setAutoGridMaker()

DelayedArray documentation built on March 25, 2021, 6:01 p.m.