NAME
PDL::CCS::Functions - Useful perl-level functions for PDL::CCS
SYNOPSIS
use PDL;
use PDL::CCS::Functions;
##---------------------------------------------------------------------
## ... stuff happens
Decoding
ccs_pointerlen
Signature: (int ptr(N+1); int len(N))
Get number of non-missing values for each axis value from a CCS-encoded offset pointer vector $ptr().
ccs_decode
Signature: (int whichnd(Ndims,Nnz); nzvals(Nnz); missing(); \@Dims; [o]a(@Dims))
Decode a CCS-encoded matrix (no dataflow).
Scalar Operations
Scalar operations can be performed in parallel directly on $nzvals
(and if applicable on $missing
as well):
$c = 42;
$nzvals2 = $nzvals + $c; $missing2 = $missing + $c;
$nzvals2 = $nzvals - $c; $missing2 = $missing - $c;
$nzvals2 = $nzvals * $c; $missing2 = $missing * $c;
$nzvals2 = $nzvals / $c; $missing2 = $missing / $c;
$nzvals2 = $nzvals ** $c; $missing2 = $missing ** $c;
$nzvals2 = log($nzvals); $missing2 = log($missing);
$nzvals2 = exp($nzvals); $missing2 = exp(missing);
$nzvals2 = $nzvals->and2($c,0); $missing2 = $missing->and($c,0);
$nzvals2 = $nzvals->or2($c,0); $missing2 = $missing->or2($c,0);
$nzvals2 = $nzvals->not(); $missing2 = $missing->not();
Nothing prevents scalar operations from producing new "missing" values (e.g. $nzvals*0), so you might want to re-encode your compressed data after applying the operation.
Vector Operations
ccs_OP_vector_mia
Signature: (int whichDimV(Nnz); nzvals(Nnz); vec(V); [o]nzvals_out(Nnz))
A number of row- and column-vector operations may be performed directly on encoded Nd-PDLs, without the need for decoding to a (potentially huge) dense temporary. These operations assume that "missing" values are annihilators with respect to the operation in question, i.e. that it holds for all $x
in $vec
that:
($missing __OP__ $x) == $missing
This is in line with the usual PDL semantics if your $missing
value is BAD
, but may produce unexpected results when e.g. adding a vector to a sparse PDL with $missing
==0. If you really need to do something like the latter, then you're probably better off decoding to a dense PDL anyway.
Predefined function names for encoded-PDL vector operations are all of the form: ccs_${OPNAME}_ma
, where ${OPNAME} is the base name of the operation:
plus ##-- addition
minus ##-- subtraction
mult ##-- multiplication (NOT matrix-multiplication)
divide ##-- division
modulo ##-- modulo
power ##-- potentiation
gt ##-- greater-than
ge ##-- greater-than-or-equal
lt ##-- less-than
le ##-- less-than-or-equal
eq ##-- equality
ne ##-- inequality
spaceship ##-- 3-way comparison
and2 ##-- binary AND
or2 ##-- binary OR
xor ##-- binary XOR
shiftleft ##-- left-shift
shiftright ##-- right-shift
\&CODE = ccs_binop_vector_mia($opName, \&PDLCODE);
Returns a generic vector-operation subroutine which reports errors as $opName
and uses \&PDLCODE to perform underlying computation.