xenial (1) PDL::Threading.1p.gz

Provided by: pdl_2.007-5_amd64 bug

NAME

       PDL::Threading - Tutorial for PDL's Threading feature

INTRODUCTION

       One of the most powerful features of PDL is threading, which can produce very compact and very fast PDL
       code by avoiding multiple nested for loops that C and BASIC users may be familiar with. The trouble is
       that it can take some getting used to, and new users may not appreciate the benefits of threading.

       Other vector based languages, such as MATLAB, use a subset of threading techniques, but PDL shines by
       completely generalizing them for all sorts of vector-based applications.

TERMINOLOGY: PIDDLE

       MATLAB typically refers to vectors, matrices, and arrays. Perl already has arrays, and the terms "vector"
       and "matrix" typically refer to one- and two-dimensional collections of data. Having no good term to
       describe their object, PDL developers coined the term "piddle" to give a name to their data type.

       A piddle consists of a series of numbers organized as an N-dimensional data set. Piddles provide
       efficient storage and fast computation of large N-dimensional matrices. They are highly optimized for
       numerical work.

THINKING IN TERMS OF THREADING

       If you have used PDL for a little while already, you may have been using threading without realising it.
       Start the PDL shell (type "perldl" or "pdl2" on a terminal). Most examples in this tutorial use the PDL
       shell.  Make sure that PDL::NiceSlice and PDL::AutoLoader are enabled. For example:

         % pdl2
         perlDL shell v1.352
         ...
         ReadLines, NiceSlice, MultiLines  enabled
        ...
         Note: AutoLoader not enabled ('use PDL::AutoLoader' recommended)

         pdl>

       In this example, NiceSlice was automatically enabled, but AutoLoader was not.  To enable it, type "use
       PDL::AutoLoader".

       Let's start with a two-dimensional piddle:

         pdl> $a = sequence(11,9)
         pdl> p $a
         [
           [ 0  1  2  3  4  5  6  7  8  9 10]
           [11 12 13 14 15 16 17 18 19 20 21]
           [22 23 24 25 26 27 28 29 30 31 32]
           [33 34 35 36 37 38 39 40 41 42 43]
           [44 45 46 47 48 49 50 51 52 53 54]
           [55 56 57 58 59 60 61 62 63 64 65]
           [66 67 68 69 70 71 72 73 74 75 76]
           [77 78 79 80 81 82 83 84 85 86 87]
           [88 89 90 91 92 93 94 95 96 97 98]
         ]

       The "info" method gives you basic information about a piddle:

         pdl> p $a->info
         PDL: Double D [11,9]

       This tells us that $a is an 11 x 9 piddle composed of double precision numbers. If we wanted to add 3 to
       all elements in an "n x m" piddle, a traditional language would use two nested for-loops:

         # Pseudo-code. Traditional way to add 3 to an array.
         for (x=0; x < n; x++) {
             for (y=0; y < m; y++) {
                 a(x,y) = a(x,y) + 3
             }
         }

       Note: Notice that indices start at 0, as in Perl, C and Java (and unlike MATLAB and IDL).

       But with PDL, we can just write:

         pdl> $b = $a + 3
         pdl> p $b
         [
           [  3   4   5   6   7   8   9  10  11  12  13]
           [ 14  15  16  17  18  19  20  21  22  23  24]
           [ 25  26  27  28  29  30  31  32  33  34  35]
           [ 36  37  38  39  40  41  42  43  44  45  46]
           [ 47  48  49  50  51  52  53  54  55  56  57]
           [ 58  59  60  61  62  63  64  65  66  67  68]
           [ 69  70  71  72  73  74  75  76  77  78  79]
           [ 80  81  82  83  84  85  86  87  88  89  90]
           [ 91  92  93  94  95  96  97  98  99 100 101]
         ]

       This is the simplest example of threading, and it is something that all numerical software tools do. The
       "+ 3" operation was automatically applied along two dimensions. Now suppose you want to to subtract a
       line from every row in $a:

         pdl> $line = sequence(11)
         pdl> p $line
         [0 1 2 3 4 5 6 7 8 9 10]
         pdl> $c = $a - $line
         pdl> p $c
         [
          [ 0  0  0  0  0  0  0  0  0  0  0]
          [11 11 11 11 11 11 11 11 11 11 11]
          [22 22 22 22 22 22 22 22 22 22 22]
          [33 33 33 33 33 33 33 33 33 33 33]
          [44 44 44 44 44 44 44 44 44 44 44]
          [55 55 55 55 55 55 55 55 55 55 55]
          [66 66 66 66 66 66 66 66 66 66 66]
          [77 77 77 77 77 77 77 77 77 77 77]
          [88 88 88 88 88 88 88 88 88 88 88]
         ]

       Two things to note here: First, the value of $a is still the same. Try "p $a" to check. Second, PDL
       automatically subtracted $line from each row in $a. Why did it do that? Let's look at the dimensions of
       $a, $line and $c:

         pdl> p $line->info  =>  PDL: Double D [11]
         pdl> p $a->info     =>  PDL: Double D [11,9]
         pdl> p $c->info     =>  PDL: Double D [11,9]

       So, both $a and $line have the same number of elements in the 0th dimension! What PDL then did was thread
       over the higher dimensions in $a and repeated the same operation 9 times to all the rows on $a. This is
       PDL threading in action.

       What if you want to subtract $line from the first line in $a only?  You can do that by specifying the
       line explicitly:

         pdl> $a(:,0) -= $line
         pdl> p $a
         [
          [ 0  0  0  0  0  0  0  0  0  0  0]
          [11 12 13 14 15 16 17 18 19 20 21]
          [22 23 24 25 26 27 28 29 30 31 32]
          [33 34 35 36 37 38 39 40 41 42 43]
          [44 45 46 47 48 49 50 51 52 53 54]
          [55 56 57 58 59 60 61 62 63 64 65]
          [66 67 68 69 70 71 72 73 74 75 76]
          [77 78 79 80 81 82 83 84 85 86 87]
          [88 89 90 91 92 93 94 95 96 97 98]
         ]

       See PDL::Indexing and PDL::NiceSlice to learn more about specifying subsets from piddles.

       The true power of threading comes when you realise that the piddle can have any number of dimensions!
       Let's make a 4 dimensional piddle:

         pdl> $piddle_4D = sequence(11,3,7,2)
         pdl> $c = $piddle_4D - $line

       Now $c is a piddle of the same dimension as $piddle_4D.

         pdl> p $piddle_4D->info  =>  PDL: Double D [11,3,7,2]
         pdl> p $c->info          =>  PDL: Double D [11,3,7,2]

       This time PDL has threaded over three higher dimensions automatically, subtracting $line all the way.

       But, maybe you don't want to subtract from the rows (dimension 0), but from the columns (dimension 1).
       How do I subtract a column of numbers from each column in $a?

         pdl> $cols = sequence(9)
         pdl> p $a->info      =>  PDL: Double D [11,9]
         pdl> p $cols->info   =>  PDL: Double D [9]

       Naturally, we can't just type "$a - $cols". The dimensions don't match:

         pdl> p $a - $cols
         PDL: PDL::Ops::minus(a,b,c): Parameter 'b'
         PDL: Mismatched implicit thread dimension 0: should be 11, is 9

       How do we tell PDL that we want to subtract from  dimension 1 instead?

MANIPULATING DIMENSIONS

       There are many PDL functions that let you rearrange the dimensions of PDL arrays. They are mostly covered
       in PDL::Slices. The three most common ones are:

        xchg
        mv
        reorder

   Method: "xchg"
       The "xchg" method "exchanges" two dimensions in a piddle:

         pdl> $a = sequence(6,7,8,9)
         pdl> $a_xchg = $a->xchg(0,3)

         pdl> p $a->info       =>  PDL: Double D [6,7,8,9]
         pdl> p $a_xchg->info  =>  PDL: Double D [9,7,8,6]
                                                  |     |
                                                  V     V
                                              (dim 0) (dim 3)

       Notice that dimensions 0 and 3 were exchanged without affecting the other dimensions. Notice also that
       "xchg" does not alter $a. The original variable $a remains untouched.

   Method: "mv"
       The "mv" method "moves" one dimension, in a piddle, shifting other dimensions as necessary.

         pdl> $a = sequence(6,7,8,9)         (dim 0)
         pdl> $a_mv = $a->mv(0,3)               |
         pdl>                                   V _____
         pdl> p $a->info     =>  PDL: Double D [6,7,8,9]
         pdl> p $a_mv->info  =>  PDL: Double D [7,8,9,6]
                                                 ----- |
                                                       V
                                                     (dim 3)

       Notice that when dimension 0 was moved to position 3, all the other dimensions had to be shifted as well.
       Notice also that "mv" does not alter $a. The original variable $a remains untouched.

   Method: "reorder"
       The "reorder" method is a generalization of the "xchg" and "mv" methods.  It "reorders" the dimensions in
       any way you specify:

         pdl> $a = sequence(6,7,8,9)
         pdl> $a_reorder = $a->reorder(3,0,2,1)
         pdl>
         pdl> p $a->info          =>  PDL: Double D [6,7,8,9]
         pdl> p $a_reorder->info  =>  PDL: Double D [9,6,8,7]
                                                     | | | |
                                                     V V v V
                                        dimensions:  0 1 2 3

       Notice what happened. When we wrote "reorder(3,0,2,1)" we instructed PDL to:

        * Put dimension 3 first.
        * Put dimension 0 next.
        * Put dimension 2 next.
        * Put dimension 1 next.

       When you use the "reorder" method, all the dimensions are shuffled. Notice that "reorder" does not alter
       $a. The original variable $a remains untouched.

GOTCHA: LINKING VS ASSIGNMENT

   Linking
       By default, piddles are linked together so that changes on one will go back and affect the original as
       well.

         pdl> $a = sequence(4,5)
         pdl> $a_xchg = $a->xchg(1,0)

       Here, $a_xchg is not a separate object. It is merely a different way of looking at $a. Any change in
       $a_xchg will appear in $a as well.

         pdl> p $a
         [
          [ 0  1  2  3]
          [ 4  5  6  7]
          [ 8  9 10 11]
          [12 13 14 15]
          [16 17 18 19]
         ]
         pdl> $a_xchg += 3
         pdl> p $a
         [
          [ 3  4  5  6]
          [ 7  8  9 10]
          [11 12 13 14]
          [15 16 17 18]
          [19 20 21 22]
         ]

   Assignment
       Some times, linking is not the behaviour you want. If you want to make the piddles independent, use the
       "copy" method:

         pdl> $a = sequence(4,5)
         pdl> $a_xchg = $a->copy->xchg(1,0)

       Now $a and $a_xchg are completely separate objects:

         pdl> p $a
         [
          [ 0  1  2  3]
          [ 4  5  6  7]
          [ 8  9 10 11]
          [12 13 14 15]
          [16 17 18 19]
         ]
         pdl> $a_xchg += 3
         pdl> p $a
         [
          [ 0  1  2  3]
          [ 4  5  6  7]
          [ 8  9 10 11]
          [12 13 14 15]
          [16 17 18 19]
         ]
         pdl> $a_xchg
         [
          [ 3  7 11 15 19]
          [ 4  8 12 16 20]
          [ 5  9 13 17 21]
          [ 6 10 14 18 22]
         ]

PUTTING IT ALL TOGETHER

       Now we are ready to solve the problem that motivated this whole discussion:

         pdl> $a = sequence(11,9)
         pdl> $cols = sequence(9)
         pdl>
         pdl> p $a->info     =>  PDL: Double D [11,9]
         pdl> p $cols->info  =>  PDL: Double D [9]

       How do we tell PDL to subtract $cols along dimension 1 instead of dimension 0?  The simplest way is to
       use the "xchg" method and rely on PDL linking:

         pdl> p $a
         [
          [ 0  1  2  3  4  5  6  7  8  9 10]
          [11 12 13 14 15 16 17 18 19 20 21]
          [22 23 24 25 26 27 28 29 30 31 32]
          [33 34 35 36 37 38 39 40 41 42 43]
          [44 45 46 47 48 49 50 51 52 53 54]
          [55 56 57 58 59 60 61 62 63 64 65]
          [66 67 68 69 70 71 72 73 74 75 76]
          [77 78 79 80 81 82 83 84 85 86 87]
          [88 89 90 91 92 93 94 95 96 97 98]
         ]
         pdl> $a->xchg(1,0) -= $cols
         pdl> p $a
         [
          [ 0  1  2  3  4  5  6  7  8  9 10]
          [10 11 12 13 14 15 16 17 18 19 20]
          [20 21 22 23 24 25 26 27 28 29 30]
          [30 31 32 33 34 35 36 37 38 39 40]
          [40 41 42 43 44 45 46 47 48 49 50]
          [50 51 52 53 54 55 56 57 58 59 60]
          [60 61 62 63 64 65 66 67 68 69 70]
          [70 71 72 73 74 75 76 77 78 79 80]
          [80 81 82 83 84 85 86 87 88 89 90]
         ]

       General Strategy:
            Move the dimensions you want to operate on to the start of your piddle's dimension list. Then let
            PDL thread over the higher dimensions.

EXAMPLE: CONWAY'S GAME OF LIFE

       Okay, enough theory. Let's do something a bit more interesting: We'll write Conway's Game of Life in PDL
       and see how powerful PDL can be!

       The Game of Life is a simulation run on a big two dimensional grid. Each cell in the grid can either be
       alive or dead (represented by 1 or 0). The next generation of cells in the grid is calculated with simple
       rules according to the number of living cells in it's immediate neighbourhood:

       1) If an empty cell has exactly three neighbours, a living cell is generated.

       2) If a living cell has less than two neighbours, it dies of overfeeding.

       3) If a living cell has 4 or more neighbours, it dies from starvation.

       Only the first generation of cells is determined by the programmer. After that, the simulation runs
       completely according to these rules. To calculate the next generation, you need to look at each cell in
       the 2D field (requiring two loops), calculate the number of live cells adjacent to this cell (requiring
       another two loops) and then fill the next generation grid.

   Classical implementation
       Here's a classic way of writing this program in Perl. We only use PDL for addressing individual cells.

         #!/usr/bin/perl -w
         use PDL;
         use PDL::NiceSlice;

         # Make a board for the game of life.
         my $nx = 20;
         my $ny = 20;

         # Current generation.
         my $a = zeroes($nx, $ny);

         # Next generation.
         my $n = zeroes($nx, $ny);

         # Put in a simple glider.
         $a(1:3,1:3) .= pdl ( [1,1,1],
                              [0,0,1],
                              [0,1,0] );

         for (my $i = 0; $i < 100; $i++) {
           $n = zeroes($nx, $ny);
           $new_a = $a->copy;
           for ($x = 0; $x < $nx; $x++) {
               for ($y = 0; $y < $ny; $y++) {

                   # For each cell, look at the surrounding neighbours.
                   for ($dx = -1; $dx <= 1; $dx++) {
                       for ($dy = -1; $dy <= 1; $dy++) {
                            $px = $x + $dx;
                            $py = $y + $dy;

                            # Wrap around at the edges.
                            if ($px < 0) {$px = $nx-1};
                            if ($py < 0) {$py = $ny-1};
                            if ($px >= $nx) {$px = 0};
                            if ($py >= $ny) {$py = 0};

                           $n($x,$y)  .= $n($x,$y) + $a($px,$py);
                       }
                   }
                   # Do not count the central cell itself.
                   $n($x,$y) -= $a($x,$y);

                   # Work out if cell lives or dies:
                   #   Dead cell lives if n = 3
                   #   Live cell dies if n is not 2 or 3
                   if ($a($x,$y) == 1) {
                       if ($n($x,$y) < 2) {$new_a($x,$y) .= 0};
                       if ($n($x,$y) > 3) {$new_a($x,$y) .= 0};
                   } else {
                       if ($n($x,$y) == 3) {$new_a($x,$y) .= 1}
                   }
               }
           }

           print $a;

           $a = $new_a;
         }

       If you run this, you will see a small glider crawl diagonally across the grid of zeroes. On my machine,
       it prints out a couple of generations per second.

   Threaded PDL implementation
       And here's the threaded version in PDL. Just four lines of PDL code, and one of those is printing out the
       latest generation!

         #!/usr/bin/perl -w
         use PDL;
         use PDL::NiceSlice;

         my $a = zeroes(20,20);

         # Put in a simple glider.
         $a(1:3,1:3) .= pdl ( [1,1,1],
                              [0,0,1],
                              [0,1,0] );

         my $n;
         for (my $i = 0; $i < 100; $i++) {
           # Calculate the number of neighbours per cell.
           $n = $a->range(ndcoords($a)-1,3,"periodic")->reorder(2,3,0,1);
           $n = $n->sumover->sumover - $a;

           # Calculate the next generation.
           $a = ((($n == 2) + ($n == 3))* $a) + (($n==3) * !$a);

           print $a;
         }

       The threaded PDL version is much faster:

         Classical => 32.79 seconds.
         Threaded  =>  0.41 seconds.

   Explanation
       How does the threaded version work?

       There are many PDL functions designed to help you carry out PDL threading.  In this example, the key
       functions are:

       Method: "range"

       At the simplest level, the "range" method is a different way to select a portion of a piddle. Instead of
       using the "$a(2,3)" notation, we use another piddle.

         pdl> $a = sequence(6,7)
         pdl> p $a
         [
          [ 0  1  2  3  4  5]
          [ 6  7  8  9 10 11]
          [12 13 14 15 16 17]
          [18 19 20 21 22 23]
          [24 25 26 27 28 29]
          [30 31 32 33 34 35]
          [36 37 38 39 40 41]
         ]
         pdl> p $a->range( pdl [1,2] )
         13
         pdl> p $a(1,2)
         [
          [13]
         ]

       At this point, the "range" method looks very similar to a regular PDL slice.  But the "range" method is
       more general. For example, you can select several components at once:

         pdl> $index = pdl [ [1,2],[2,3],[3,4],[4,5] ]
         pdl> p $a->range( $index )
         [13 20 27 34]

       Additionally, "range" takes a second parameter which determines the size of the chunk to return:

         pdl> $size = 3
         pdl> p $a->range( pdl([1,2]) , $size )
         [
          [13 14 15]
          [19 20 21]
          [25 26 27]
         ]

       We can use this to select one or more 3x3 boxes.

       Finally, "range" can take a third parameter called the "boundary" condition.  It tells PDL what to do if
       the size box you request goes beyond the edge of the piddle. We won't go over all the options. We'll just
       say that the option "periodic" means that the piddle "wraps around". For example:

         pdl> p $a
         [
          [ 0  1  2  3  4  5]
          [ 6  7  8  9 10 11]
          [12 13 14 15 16 17]
          [18 19 20 21 22 23]
          [24 25 26 27 28 29]
          [30 31 32 33 34 35]
          [36 37 38 39 40 41]
         ]
         pdl> $size = 3
         pdl> p $a->range( pdl([4,2]) , $size , "periodic" )
         [
          [16 17 12]
          [22 23 18]
          [28 29 24]
         ]
         pdl> p $a->range( pdl([5,2]) , $size , "periodic" )
         [
          [17 12 13]
          [23 18 19]
          [29 24 25]
         ]

       Notice how the box wraps around the boundary of the piddle.

       Method: "ndcoords"

       The "ndcoords" method is a convenience method that returns an enumerated list of coordinates suitable for
       use with the "range" method.

         pdl> p $piddle = sequence(3,3)
         [
          [0 1 2]
          [3 4 5]
          [6 7 8]
         ]
         pdl> p ndcoords($piddle)
         [
          [
           [0 0]
           [1 0]
           [2 0]
          ]
          [
           [0 1]
           [1 1]
           [2 1]
          ]
          [
           [0 2]
           [1 2]
           [2 2]
          ]
         ]

       This can be a little hard to read. Basically it's saying that the coordinates for every element in
       $piddle is given by:

          (0,0)     (1,0)     (2,0)
          (1,0)     (1,1)     (2,1)
          (2,0)     (2,1)     (2,2)

       Combining "range" and "ndcoords"

       What really matters is that "ndcoords" is designed to work together with "range", with no $size
       parameter, you get the same piddle back.

         pdl> p $piddle
         [
          [0 1 2]
          [3 4 5]
          [6 7 8]
         ]
         pdl> p $piddle->range( ndcoords($piddle) )
         [
          [0 1 2]
          [3 4 5]
          [6 7 8]
         ]

       Why would this be useful? Because now we can ask for a series of "boxes" for the entire piddle. For
       example, 2x2 boxes:

         pdl> p $piddle->range( ndcoords($piddle) , 2 , "periodic" )

       The output of this function is difficult to read because the "boxes" along the last two dimension. We can
       make the result more readable by rearranging the dimensions:

         pdl> p $piddle->range( ndcoords($piddle) , 2 , "periodic" )->reorder(2,3,0,1)
         [
          [
           [
            [0 1]
            [3 4]
           ]
           [
            [1 2]
            [4 5]
           ]
           ...
         ]

       Here you can see more clearly that

         [0 1]
         [3 4]

       Is the 2x2 box starting with the (0,0) element of $piddle.

       We are not done yet. For the game of life, we want 3x3 boxes from $a:

         pdl> p $a
         [
          [ 0  1  2  3  4  5]
          [ 6  7  8  9 10 11]
          [12 13 14 15 16 17]
          [18 19 20 21 22 23]
          [24 25 26 27 28 29]
          [30 31 32 33 34 35]
          [36 37 38 39 40 41]
         ]
         pdl> p $a->range( ndcoords($a) , 3 , "periodic" )->reorder(2,3,0,1)
         [
          [
           [
            [ 0  1  2]
            [ 6  7  8]
            [12 13 14]
           ]
           ...
         ]

       We can confirm that this is the 3x3 box starting with the (0,0) element of $a.  But there is one problem.
       We actually want the 3x3 box to be centered on (0,0). That's not a problem. Just subtract 1 from all the
       coordinates in "ndcoords($a)". Remember that the "periodic" option takes care of making everything wrap
       around.

         pdl> p $a->range( ndcoords($a) - 1 , 3 , "periodic" )->reorder(2,3,0,1)
         [
          [
           [
            [41 36 37]
            [ 5  0  1]
            [11  6  7]
           ]
           [
            [36 37 38]
            [ 0  1  2]
            [ 6  7  8]
           ]
           ...

       Now we see a 3x3 box with the (0,0) element in the centre of the box.

       Method: "sumover"

       The "sumover" method adds along only the first dimension. If we apply it twice, we will be adding all the
       elements of each 3x3 box.

         pdl> $n = $a->range(ndcoords($a)-1,3,"periodic")->reorder(2,3,0,1)
         pdl> p $n
         [
          [
           [
            [41 36 37]
            [ 5  0  1]
            [11  6  7]
           ]
           [
            [36 37 38]
            [ 0  1  2]
            [ 6  7  8]
           ]
           ...
         pdl> p $n->sumover->sumover
         [
          [144 135 144 153 162 153]
          [ 72  63  72  81  90  81]
          [126 117 126 135 144 135]
          [180 171 180 189 198 189]
          [234 225 234 243 252 243]
          [288 279 288 297 306 297]
          [216 207 216 225 234 225]
         ]

       Use a calculator to confirm that 144 is the sum of all the elements in the first 3x3 box and 135 is the
       sum of all the elements in the second 3x3 box.

       Counting neighbours

       We are almost there!

       Adding up all the elements in a 3x3 box is not quite what we want. We don't want to count the center box.
       Fortunately, this is an easy fix:

         pdl> p $n->sumover->sumover - $a
         [
          [144 134 142 150 158 148]
          [ 66  56  64  72  80  70]
          [114 104 112 120 128 118]
          [162 152 160 168 176 166]
          [210 200 208 216 224 214]
          [258 248 256 264 272 262]
          [180 170 178 186 194 184]
         ]

       When applied to Conway's Game of Life, this will tell us how many living neighbours each cell has:

         pdl> $a = zeroes(10,10)
         pdl> $a(1:3,1:3) .= pdl ( [1,1,1],
         ..(    >                  [0,0,1],
         ..(    >                  [0,1,0] )
         pdl> p $a
         [
          [0 0 0 0 0 0 0 0 0 0]
          [0 1 1 1 0 0 0 0 0 0]
          [0 0 0 1 0 0 0 0 0 0]
          [0 0 1 0 0 0 0 0 0 0]
          [0 0 0 0 0 0 0 0 0 0]
          [0 0 0 0 0 0 0 0 0 0]
          [0 0 0 0 0 0 0 0 0 0]
          [0 0 0 0 0 0 0 0 0 0]
          [0 0 0 0 0 0 0 0 0 0]
          [0 0 0 0 0 0 0 0 0 0]
         ]
         pdl> $n = $a->range(ndcoords($a)-1,3,"periodic")->reorder(2,3,0,1)
         pdl> $n = $n->sumover->sumover - $a
         pdl> p $n
         [
          [1 2 3 2 1 0 0 0 0 0]
          [1 1 3 2 2 0 0 0 0 0]
          [1 3 5 3 2 0 0 0 0 0]
          [0 1 1 2 1 0 0 0 0 0]
          [0 1 1 1 0 0 0 0 0 0]
          [0 0 0 0 0 0 0 0 0 0]
          [0 0 0 0 0 0 0 0 0 0]
          [0 0 0 0 0 0 0 0 0 0]
          [0 0 0 0 0 0 0 0 0 0]
          [0 0 0 0 0 0 0 0 0 0]
         ]

       For example, this tells us that cell (0,0) has 1 living neighbour, while cell (2,2) has 5 living
       neighbours.

       Calculating the next generation

       At this point, the variable $n has the number of living neighbours for every cell. Now we apply the rules
       of the game of life to calculate the next generation.

       If an empty cell has exactly three neighbours, a living cell is generated.
            Get a list of cells that have exactly three neighbours:

              pdl> p ($n == 3)
              [
               [0 0 1 0 0 0 0 0 0 0]
               [0 0 1 0 0 0 0 0 0 0]
               [0 1 0 1 0 0 0 0 0 0]
               [0 0 0 0 0 0 0 0 0 0]
               [0 0 0 0 0 0 0 0 0 0]
               [0 0 0 0 0 0 0 0 0 0]
               [0 0 0 0 0 0 0 0 0 0]
               [0 0 0 0 0 0 0 0 0 0]
               [0 0 0 0 0 0 0 0 0 0]
               [0 0 0 0 0 0 0 0 0 0]
              ]

            Get a list of empty cells that have exactly three neighbours:

              pdl> p ($n == 3) * !$a

       If a living cell has less than 2 or more than 3 neighbours, it dies.
            Get a list of cells that have exactly 2 or 3 neighbours:

              pdl> p (($n == 2) + ($n == 3))
              [
               [0 1 1 1 0 0 0 0 0 0]
               [0 0 1 1 1 0 0 0 0 0]
               [0 1 0 1 1 0 0 0 0 0]
               [0 0 0 1 0 0 0 0 0 0]
               [0 0 0 0 0 0 0 0 0 0]
               [0 0 0 0 0 0 0 0 0 0]
               [0 0 0 0 0 0 0 0 0 0]
               [0 0 0 0 0 0 0 0 0 0]
               [0 0 0 0 0 0 0 0 0 0]
               [0 0 0 0 0 0 0 0 0 0]
              ]

            Get a list of living cells that have exactly 2 or 3 neighbours:

              pdl> p (($n == 2) + ($n == 3)) * $a

       Putting it all together, the next generation is:

         pdl> $a = ((($n == 2) + ($n == 3)) * $a) + (($n == 3) * !$a)
         pdl> p $a
         [
          [0 0 1 0 0 0 0 0 0 0]
          [0 0 1 1 0 0 0 0 0 0]
          [0 1 0 1 0 0 0 0 0 0]
          [0 0 0 0 0 0 0 0 0 0]
          [0 0 0 0 0 0 0 0 0 0]
          [0 0 0 0 0 0 0 0 0 0]
          [0 0 0 0 0 0 0 0 0 0]
          [0 0 0 0 0 0 0 0 0 0]
          [0 0 0 0 0 0 0 0 0 0]
          [0 0 0 0 0 0 0 0 0 0]
         ]

   Bonus feature: Graphics!
       If you have PDL::Graphics::TriD installed, you can make a graphical version of the program by just
       changing three lines:

         #!/usr/bin/perl
         use PDL;
         use PDL::NiceSlice;
         use PDL::Graphics::TriD;

         my $a = zeroes(20,20);

         # Put in a simple glider.
         $a(1:3,1:3) .= pdl ( [1,1,1],
                              [0,0,1],
                              [0,1,0] );

         my $n;
         for (my $i = 0; $i < 100; $i++) {
             # Calculate the number of neighbours per cell.
             $n = $a->range(ndcoords($a)-1,3,"periodic")->reorder(2,3,0,1);
             $n = $n->sumover->sumover - $a;

             # Calculate the next generation.
             $a = ((($n == 2) + ($n == 3))* $a) + (($n==3) * !$a);

             # Display.
             nokeeptwiddling3d();
             imagrgb [$a];
         }

       But if we really want to see something interesting, we should make a few more changes:

       1) Start with a random collection of 1's and 0's.

       2) Make the grid larger.

       3) Add a small timeout so we can see the game evolve better.

       4) Use a while loop so that the program can run as long as it needs to.

         #!/usr/bin/perl
         use PDL;
         use PDL::NiceSlice;
         use PDL::Graphics::TriD;
         use Time::HiRes qw(usleep);

         my $a = random(100,100);
         $a = ($a < 0.5);

         my $n;
         while (1) {
             # Calculate the number of neighbours per cell.
             $n = $a->range(ndcoords($a)-1,3,"periodic")->reorder(2,3,0,1);
             $n = $n->sumover->sumover - $a;

             # Calculate the next generation.
             $a = ((($n == 2) + ($n == 3))* $a) + (($n==3) * !$a);

             # Display.
             nokeeptwiddling3d();
             imagrgb [$a];

             # Sleep for 0.1 seconds.
             usleep(100000);
         }

CONCLUSION: GENERAL STRATEGY

       The general strategy is: Move the dimensions you want to operate on to the start of your piddle's
       dimension list. Then let PDL thread over the higher dimensions.

       Threading is a powerful tool that helps eliminate for-loops and can make your code more concise.
       Hopefully this tutorial has shown why it is worth getting to grips with threading in PDL.

       Copyright 2010 Matthew Kenworthy (kenworthy@strw.leidenuniv.nl) and Daniel Carrera (dcarrera@gmail.com).
       You can distribute and/or modify this document under the same terms as the current Perl license.

       See: http://dev.perl.org/licenses/