PDL::Broadcasting - Tutorial for PDL's Broadcasting feature
One of the most powerful features of PDL is
broadcasting, 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
broadcasting.
Other vector based languages, such as MATLAB, use a subset of broadcasting
techniques, but PDL shines by completely generalizing them for all sorts of
vector-based applications.
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 "
ndarray" to give a name to their data type.
An
ndarray consists of a series of numbers organized as an N-dimensional
data set. ndarrays provide efficient storage and fast computation of large
N-dimensional matrices. They are highly optimized for numerical work.
If you have used PDL for a little while already, you may have been using
broadcasting 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
ndarray:
pdl> $x = sequence(11,9)
pdl> p $x
[
[ 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 an
ndarray:
pdl> p $x->info
PDL: Double D [11,9]
This tells us that $x is an 11 x 9
ndarray composed of double precision
numbers. If we wanted to add 3 to all elements in an "n x m"
ndarray, a traditional language would use two nested for-loops:
# Pseudo-code. Traditional way to add 3 to an array.
for (i=0; i < n; i++) {
for (j=0; j < m; j++) {
a(i,j) = a(i,j) + 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> $y = $x + 3
pdl> p $y
[
[ 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 broadcasting, 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 $x:
pdl> $line = sequence(11)
pdl> p $line
[0 1 2 3 4 5 6 7 8 9 10]
pdl> $c = $x - $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 $x is still the same. Try "p
$x" to check. Second, PDL automatically subtracted $line from each row in
$x. Why did it do that? Let's look at the dimensions of $x, $line and $c:
pdl> p $line->info => PDL: Double D [11]
pdl> p $x->info => PDL: Double D [11,9]
pdl> p $c->info => PDL: Double D [11,9]
So, both $x and $line have the same number of elements in the 0th dimension!
What PDL then did was broadcast over the higher dimensions in $x and repeated
the same operation 9 times to all the rows on $x. This is PDL broadcasting in
action.
What if you want to subtract $line from the first line in $x only? You can do
that by specifying the line explicitly:
pdl> $x(:,0) -= $line
pdl> p $x
[
[ 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
ndarrays.
The true power of broadcasting comes when you realise that the ndarray can have
any number of dimensions! Let's make a 4 dimensional ndarray:
pdl> $ndarray_4D = sequence(11,3,7,2)
pdl> $c = $ndarray_4D - $line
Now $c is an ndarray of the same dimension as $ndarray_4D.
pdl> p $ndarray_4D->info => PDL: Double D [11,3,7,2]
pdl> p $c->info => PDL: Double D [11,3,7,2]
This time PDL has broadcasted 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 $x?
pdl> $cols = sequence(9)
pdl> p $x->info => PDL: Double D [11,9]
pdl> p $cols->info => PDL: Double D [9]
Naturally, we can't just type "$x - $cols". The dimensions don't
match:
pdl> p $x - $cols
PDL: PDL::Ops::minus(a,b,c): Parameter 'b'
PDL: Mismatched implicit broadcast dimension 0: should be 11, is 9
How do we tell PDL that we want to subtract from dimension 1 instead?
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
The "xchg" method "
exchanges" two dimensions in an
ndarray:
pdl> $x = sequence(6,7,8,9)
pdl> $x_xchg = $x->xchg(0,3)
pdl> p $x->info => PDL: Double D [6,7,8,9]
pdl> p $x_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 $x. The original
variable $x remains untouched.
The "mv" method "
moves" one dimension, in an
ndarray, shifting other dimensions as necessary.
pdl> $x = sequence(6,7,8,9) (dim 0)
pdl> $x_mv = $x->mv(0,3) |
pdl> V _____
pdl> p $x->info => PDL: Double D [6,7,8,9]
pdl> p $x_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 $x.
The original variable $x remains untouched.
The "reorder" method is a generalization of the "xchg" and
"mv" methods. It "
reorders" the dimensions in any
way you specify:
pdl> $x = sequence(6,7,8,9)
pdl> $x_reorder = $x->reorder(3,0,2,1)
pdl>
pdl> p $x->info => PDL: Double D [6,7,8,9]
pdl> p $x_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 $x. The original variable $x
remains untouched.
By default, ndarrays are
linked together so that changes on one will go
back and affect the original
as well.
pdl> $x = sequence(4,5)
pdl> $x_xchg = $x->xchg(1,0)
Here, $x_xchg
is not a separate object. It is merely a different way of
looking at $x. Any change in $x_xchg will appear in $x as well.
pdl> p $x
[
[ 0 1 2 3]
[ 4 5 6 7]
[ 8 9 10 11]
[12 13 14 15]
[16 17 18 19]
]
pdl> $x_xchg += 3
pdl> p $x
[
[ 3 4 5 6]
[ 7 8 9 10]
[11 12 13 14]
[15 16 17 18]
[19 20 21 22]
]
Some times, linking is not the behaviour you want. If you want to make the
ndarrays independent, use the "copy" method:
pdl> $x = sequence(4,5)
pdl> $x_xchg = $x->copy->xchg(1,0)
Now $x and $x_xchg are completely separate objects:
pdl> p $x
[
[ 0 1 2 3]
[ 4 5 6 7]
[ 8 9 10 11]
[12 13 14 15]
[16 17 18 19]
]
pdl> $x_xchg += 3
pdl> p $x
[
[ 0 1 2 3]
[ 4 5 6 7]
[ 8 9 10 11]
[12 13 14 15]
[16 17 18 19]
]
pdl> $x_xchg
[
[ 3 7 11 15 19]
[ 4 8 12 16 20]
[ 5 9 13 17 21]
[ 6 10 14 18 22]
]
Now we are ready to solve the problem that motivated this whole discussion:
pdl> $x = sequence(11,9)
pdl> $cols = sequence(9)
pdl>
pdl> p $x->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 $x
[
[ 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> $x->xchg(1,0) -= $cols
pdl> p $x
[
[ 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 ndarray's dimension list. Then let PDL broadcast over the higher
dimensions.
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.
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 $a1 = zeroes($nx, $ny);
# Next generation.
my $n = zeroes($nx, $ny);
# Put in a simple glider.
$a1(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 = $a1->copy;
for ($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) + $a1($px,$py);
}
}
# Do not count the central cell itself.
$n($x,$y) -= $a1($x,$y);
# Work out if cell lives or dies:
# Dead cell lives if n = 3
# Live cell dies if n is not 2 or 3
if ($a1($x,$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 $a1;
$a1 = $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.
And here's the broadcasted 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 $x = zeroes(20,20);
# Put in a simple glider.
$x(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 = $x->range(ndcoords($x)-1,3,"periodic")->reorder(2,3,0,1);
$n = $n->sumover->sumover - $x;
# Calculate the next generation.
$x = ((($n == 2) + ($n == 3))* $x) + (($n==3) * !$x);
print $x;
}
The broadcasted PDL version is much faster:
Classical => 32.79 seconds.
Broadcasting => 0.41 seconds.
How does the broadcasted version work?
There are many PDL functions designed to help you carry out PDL broadcasting. In
this example, the key functions are:
Method: "range"
At the simplest level, the "range" method is a different way to select
a portion of an ndarray. Instead of using the "$x(2,3)" notation, we
use another ndarray.
pdl> $x = sequence(6,7)
pdl> p $x
[
[ 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 $x->range( pdl [1,2] )
13
pdl> p $x(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 $x->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 $x->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 ndarray. We won't go over all the options.
We'll just say that the option "periodic" means that the ndarray
"wraps around". For example:
pdl> p $x
[
[ 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 $x->range( pdl([4,2]) , $size , "periodic" )
[
[16 17 12]
[22 23 18]
[28 29 24]
]
pdl> p $x->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 ndarray.
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 $ndarray = sequence(3,3)
[
[0 1 2]
[3 4 5]
[6 7 8]
]
pdl> p ndcoords($ndarray)
[
[
[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 $ndarray 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 ndarray
back.
pdl> p $ndarray
[
[0 1 2]
[3 4 5]
[6 7 8]
]
pdl> p $ndarray->range( ndcoords($ndarray) )
[
[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 ndarray. For example, 2x2 boxes:
pdl> p $ndarray->range( ndcoords($ndarray) , 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 $ndarray->range( ndcoords($ndarray) , 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 $ndarray.
We are not done yet. For the game of life, we want 3x3 boxes from $x:
pdl> p $x
[
[ 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 $x->range( ndcoords($x) , 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 $x.
But there is one problem. We actually want the 3x3 box to be
centered
on (0,0). That's not a problem. Just subtract 1 from all the coordinates in
"ndcoords($x)". Remember that the "periodic" option takes
care of making everything wrap around.
pdl> p $x->range( ndcoords($x) - 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 = $x->range(ndcoords($x)-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 - $x
[
[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> $x = zeroes(10,10)
pdl> $x(1:3,1:3) .= pdl ( [1,1,1],
..( > [0,0,1],
..( > [0,1,0] )
pdl> p $x
[
[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 = $x->range(ndcoords($x)-1,3,"periodic")->reorder(2,3,0,1)
pdl> $n = $n->sumover->sumover - $x
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 ($n == 3) * !$x
- 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)) * $x
Putting it all together, the next generation is:
pdl> $x = ((($n == 2) + ($n == 3)) * $x) + (($n == 3) * !$x)
pdl> p $x
[
[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]
]
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 $x = zeroes(20,20);
# Put in a simple glider.
$x(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 = $x->range(ndcoords($x)-1,3,"periodic")->reorder(2,3,0,1);
$n = $n->sumover->sumover - $x;
# Calculate the next generation.
$x = ((($n == 2) + ($n == 3))* $x) + (($n==3) * !$x);
# Display.
nokeeptwiddling3d();
imagrgb [$x];
}
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 $x = random(100,100);
$x = ($x < 0.5);
my $n;
while (1) {
# Calculate the number of neighbours per cell.
$n = $x->range(ndcoords($x)-1,3,"periodic")->reorder(2,3,0,1);
$n = $n->sumover->sumover - $x;
# Calculate the next generation.
$x = ((($n == 2) + ($n == 3))* $x) + (($n==3) * !$x);
# Display.
nokeeptwiddling3d();
imagrgb [$x];
# Sleep for 0.1 seconds.
usleep(100000);
}
The general strategy is:
Move the dimensions you want to operate on to
the start of your ndarray's dimension list. Then let PDL broadcast over
the higher dimensions.
Broadcasting 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 broadcasting in PDL.
Copyright 2010 Matthew Kenworthy (
[email protected]) and Daniel
Carrera (
[email protected]). You can distribute and/or modify this document
under the same terms as the current Perl license.
See:
http://dev.perl.org/licenses/