Running mean is simple — just use convolution. I do things like this:
$runningmean = $source->convolveND( ones($how_wide)/$how_wide,
{bound=>’c’} );
Running median is harder since it’s not a linear operation and the median must
be computed at each point. Since, like most interpreted vectorized languages,
PDL
is pessimal for memory access, you can do *much* better wth a C code. I don’t
know
of a 1-D running median operator in the distribution itself. However, there’s
a
2-D running median, which you could use:
$runningmedian =
$source->(:,*1)->med2d(ones($how_wide,1))->(:,(0))->sever;
That uses stupid index tricks to treat your 1-D array as a 2-D array, by giving
it a dummy dimension in position 1. It’s not as efficient as a PP (compiled)
running-
median operator could be, but it’s far more efficient than the threading
solution
(since it will make good use of CPU cache unless $how_wide is enormous).
Cheers,
Craig
> On Mar 1, 2016, at 9:06 AM, Pakhomov Yury <pakho...@inasan.ru> wrote:
>
> Hello all!
>
> How to write PDL efficient code for computing the moving average or
> moving median?
>
> For example, I want to use the median filter width of 33 pixels for data:
>
> $x=sequence(1000);
> $y=sin($x/10)+random($x)/10;
>
> For data of little size I can create new piddle size of (33,nelem($y))
> where each row contains a portion of data size of 33 which then I
> include in "medover":
>
> $med=&medianPDL($y,33);
>
> sub medianPDL{
> my $pdl=shift;
> my $win=shift;
> my $tmp=zeroes($win,nelem($pdl));
> my $inds=yvals($tmp)-int($win/2)+xvals($tmp);
> my $a=$inds->minimum;
> $a->where($a>=0).=0;
> $inds-=$a->dummy(0,$win);
> my $b=$inds->maximum-$tmp->getdim(1)+1;
> $b->where($b<0).=0;
> $inds-=$b->dummy(0,$win);
>
> return $pdl->index($inds)->medover;
> }
>
> ________________________________
>
> May be exists a more efficient way to compute the moving average or
> moving median?
>
> Yury
>
>
> ------------------------------------------------------------------------------
> Site24x7 APM Insight: Get Deep Visibility into Application Performance
> APM + Mobile APM + RUM: Monitor 3 App instances at just $35/Month
> Monitor end-to-end web transactions and take corrective actions now
> Troubleshoot faster and improve end-user experience. Signup Now!
> http://pubads.g.doubleclick.net/gampad/clk?id=272487151&iu=/4140
> _______________________________________________
> pdl-general mailing list
> pdl-general@lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/pdl-general
>
------------------------------------------------------------------------------
Site24x7 APM Insight: Get Deep Visibility into Application Performance
APM + Mobile APM + RUM: Monitor 3 App instances at just $35/Month
Monitor end-to-end web transactions and take corrective actions now
Troubleshoot faster and improve end-user experience. Signup Now!
http://pubads.g.doubleclick.net/gampad/clk?id=272487151&iu=/4140
_______________________________________________
pdl-general mailing list
pdl-general@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/pdl-general
