DimSqueezer_CL: Class '"DimSqueezer_CL"'

In SOD: SOD for multidimensional scaling.

Description

Provides a reference class that implements the self-organising-deltoids (SOD) multi-dimensional-scaling algorithm (MDS) using openCL to provide parallel execution on a general purpose GPU.
New objects are made by calling the initializer using either DimSqueezer_CL$new(m) or new("DimSqueezer_CL", m).
The initializer returns an object from which the methods and fields are accessed using the $ operator. See the examples section for usage.

Extends

All reference classes extend and inherit methods from "envRefClass".

Warning

• This class will only be available on computers where openCL functionality is available.
  DANGER: On Mac OSX, it seems possible to crash the display driver (i.e. the computer is unusable, though not in a completely hung state) if specifying a work group size (wksize argument) that is larger than the PREFERRED_WORK_GROUP_SIZE_MULTIPLE. (This value is printed when the class is initialized.)
  Binary distributions of this package are unlikely to provide access this class, though that will depend on the compilation environment. I have only tested this on a small number of computers, and am not sure how it will behave on other systems. Precompiled binaries for Windows will by default not have access to the openCL functionality.

• This is a reference class that does not follow normal R semantic rules. Copying objects does not make new copies, but additional references. See ?setRefClass for more details.

• This class keeps an external pointer to a C++ object. This object will be destroyed at the end of a session. If you wish to use it after reloading a session you will need to recreate the object. This can be done using the data.matrix field of the object (for a DimSqueezer_CL object called ds)
  ds <- DimSqueezer_CL$new( ds$data.matrix )

Fields

pointer:

Object of class externalptr that refers to the C++ DimSqueezer_CL object (read only).

data.matrix:

Object of class matrix a numeric matrix used to create the DimSqueezer.

Methods

squeeze(target_dim, iter_no, wksize):

Squeezes points into target_dim dimensions in iter_no iterations using an openCL work group size of wksize (see Warning). Returns a named list of class sod_sq3 containing the new coordinates in $pos, the stress evolution in $stress, the mapping dimensionality in mapDims and the stress (sum error) at each node in node_stress.

initialize(m):

Initializes the object. m should be a numeric matrix (use by calling new as in examples).

Author(s)

Martin Jakt

References

http://martin.jakt.org.uk/

See Also

DimSqueezer
serialDimFactors
parallelDimFactors
parallelExpDimFactors
sod_sq3

Examples

## Not run: 
showClass("DimSqueezer_CL")

## Make an openCL accelerated squeezer object
## note this is only available where both the physical
## hardware and the appropriate development environment
## is available.
data(f186)

## use columns 7:12 as these have comparable data
ds <- DimSqueezer_CL$new( as.matrix(f186[,7:12]) )

## squeeze, 1000 iterations using a work group size of 64
## giving a 2 dimensional representation.
sq <- ds$squeeze(2, 200, 64)

## functions to plot the relationships
plotPoints(sq)

## use columns of f186 to make a more interesting
## plot
plotPoints(sq, col=hsvScale(f186[r,"Etv2"]), pch=19)
plotPoints(sq, col=hsvScale(f186[r,"p_Etv2"]), pch=19)

## and a funky plot
plotConcentric(sq, f186[r,7:12], cex.max=5, pch=19, leg.pos="topleft")

## and to plot the stress mapping.
plotStress(sq1)

## End(Not run)

SOD documentation built on May 30, 2017, 2:03 a.m.