| 1 | =head1 NAME | ||
| 2 | |||
| 3 | PDL::Filter::Linear - linear filtering for PDL | ||
| 4 | |||
| 5 | =head1 SYNOPSIS | ||
| 6 | |||
| 7 | $x = PDL::Filter::Linear->new( | ||
| 8 | {Weights => $v, | ||
| 9 | Point => 10}); | ||
| 10 | |||
| 11 | $y = PDL::Filter::Gaussian->new(15,2); # 15 points, 2 std devn. | ||
| 12 | |||
| 13 | ($pred,$corrslic) = $x->predict($dat); | ||
| 14 | |||
| 15 | =head1 DESCRIPTION | ||
| 16 | |||
| 17 | A wrapper for generic linear filters. | ||
| 18 | Just for convenience. This should in the future use DataPresenter. | ||
| 19 | |||
| 20 | Also, this class should at some point learn to do FFT whenever it is | ||
| 21 | useful. | ||
| 22 | |||
| 23 | =cut | ||
| 24 | |||
| 25 | package PDL::Filter::Linear; | ||
| 26 | use strict; | ||
| 27 | use warnings; | ||
| 28 | use PDL; | ||
| 29 | use PDL::Basic; | ||
| 30 | use PDL::Slices; | ||
| 31 | use PDL::Primitive; | ||
| 32 | |||
| 33 | sub new($$) { | ||
| 34 | my($type,$pars) = @_; | ||
| 35 | |||
| 36 | my $this = bless {},$type; | ||
| 37 | barf("Must specify weights\n") unless defined $pars->{Weights}; | ||
| 38 | $this->{Weights} = delete $pars->{Weights}; | ||
| 39 | $this->{Point} = defined $pars->{Point} ? $pars->{Point} : 0; | ||
| 40 | $this; | ||
| 41 | } | ||
| 42 | |||
| 43 | sub predict($$) { | ||
| 44 | my($this,$data) = @_; | ||
| 45 | my $ldata = $data->lags(0,1,$this->{Weights}->getdim(0)); | ||
| 46 | inner($ldata->transpose,$this->{Weights}, | ||
| 47 | (my $pred = PDL->null)); | ||
| 48 | return wantarray ? ($pred,$ldata->slice(":,($this->{Point})")) : | ||
| 49 | $pred ; | ||
| 50 | } | ||
| 51 | |||
| 52 | package PDL::Filter::Gaussian; | ||
| 53 | use PDL; use PDL::Basic; use PDL::Slices; use PDL::Primitive; | ||
| 54 | use strict; | ||
| 55 | |||
| 56 | @PDL::Filter::Gaussian::ISA = qw/PDL::Filter::Linear/; | ||
| 57 | |||
| 58 | sub new($$) { | ||
| 59 | my($type,$npoints,$sigma) = @_; | ||
| 60 | my $cent = int($npoints/2); | ||
| 61 | my $x = PDL->zeroes(float, $npoints )->xvals - $cent; | ||
| 62 | my $y = exp(-($x**2)/(2*$sigma**2)); | ||
| 63 | # Normalize to unit total | ||
| 64 | $y /= sum($y); | ||
| 65 | return PDL::Filter::Linear::new($type,{Weights => $y, | ||
| 66 | Point => $cent}); | ||
| 67 | } | ||
| 68 | |||
| 69 | # Savitzky-Golay (see Numerical Recipes) | ||
| 70 | package PDL::Filter::SavGol; | ||
| 71 | use PDL; use PDL::Basic; use PDL::Slices; use PDL::Primitive; | ||
| 72 | use strict; | ||
| 73 | |||
| 74 | @PDL::Filter::Gaussian::ISA = qw/PDL::Filter::Linear/; | ||
| 75 | |||
| 76 | # XXX Doesn't work | ||
| 77 | sub new($$) { | ||
| 78 | my($type,$deg,$nleft,$nright) = @_; | ||
| 79 | my $npoints = $nright + $nleft + 1; | ||
| 80 | my $x = PDL->zeroes(float, $npoints )->xvals - $nleft; | ||
| 81 | my $mat1 = PDL->zeroes(float, $npoints,$deg+1)->xvals; | ||
| 82 | for(0..$deg-1) { | ||
| 83 | (my $tmp = $mat1->slice(":,($_)")) .= ($x ** $_); | ||
| 84 | } | ||
| 85 | my $y; | ||
| 86 | # Normalize to unit total | ||
| 87 | return PDL::Filter::Linear::new($type,{Weights => $y, | ||
| 88 | Point => $nleft}); | ||
| 89 | } | ||
| 90 | |||
| 91 | 1; |