PDL::Stats(3pm)

💖 NAME 💖

PDL::Stats — a whole collection of statistics modules in Perl Data Language, plus a quick-start guide for the non-PDL girlies. No PDL experience required, bestie. ✨

---

🌸 DESCRIPTION 🌸

It loads all the modules named down below, which means their functions show up right there in your current namespace. Convenient!! 💅

Properly formatted documentations online at http://pdl-stats.sf.net 🦋

---

✨ SYNOPSIS ✨

use PDL::LiteF;        # loads fewer modules
use PDL::Stats;

# Is equivalent to the following:
use PDL::Stats::Basic;
use PDL::Stats::GLM;
use PDL::Stats::Kmeans;
use PDL::Stats::TS;

# and the following if installed;
use PDL::Stats::Distr;
use PDL::GSL::CDF;

---

🎀 QUICK-START FOR NON-PDL PEOPLE 🎀

Okay so the whole point here: enjoy PDL::Stats without having to dive face-first into PDL. You just gotta wet your feet a little, that's it. Three key words, two concepts, and one lil icing on the cake — and you're basically well on your way. 🫶

🔮 pdl

pdl is the magic word that puts PDL::Stats at your disposal. It creates a PDL numeric data object (a pdl, or ndarray) out of a perl array or array ref. And then — boom — all PDL::Stats methods can be called from the data object. (Unless the method was made for a regular perl array, obviously. 😌)

my @y = 0..5;

my $y = pdl @y;

# a simple function

my $stdv = $y->stdv;

# you can skip the intermediate $y

my $stdv = stdv( pdl @y );

# a more complex method, skipping intermediate $y

my @x1 = qw( y y y n n n );
my @x2 = qw( 1 0 1 0 1 0 )

# do a two-way analysis of variance with y as DV and x1 x2 as IVs

my %result = pdl(@y)->anova( \@x1, \@x2 );
print "$_\t$result{$_}\n" for sort keys %result;

Now if you've got a list of lists, i.e. an array of array refs, pdl will make you a multi-dimensional data object. Iconic.

my @a = ( [1,2,3,4], [0,1,2,3], [4,5,6,7] );

my $a = pdl @a;

print $a . $a->info;

# here's what you will get

[
 [1 2 3 4]
 [0 1 2 3]
 [4 5 6 7]
]
PDL: Double D [4,3]

Important vibe check: PDL::Stats puts observations in the first dimension and variables in the second dimension — so like pdl [obs, var]. Which means the example right above her is 4 observations on 3 variables. Slay.

# you can do all kinds of fancy stuff on such a 2D pdl.

my %result = $a->kmeans( {NCLUS=>2} );
print "$_\t$result{$_}\n" for sort keys %result;

⚠️ Make sure that array of array refs is rectangular, bestie. If the array refs are different sizes, pdl will pad it out with 0s to match the longest list. Not a vibe, but it's what happens.

🧚 info

info tells you the data type (yes pdls are typed, but you honestly shouldn't have to worry about it here*) and the dimensionality of the pdl — you literally saw it up in that example. Keeping track of a pdl's dimensionality is a huge help for my sanity, ngl. And like I said above: PDL::Stats uses 2D pdls with observation × variable dimensionality. Track it. 💅

*pdl uses double precision by default. But if you're working with things like epoch time, you should probably use pdl(long, @epoch) to maintain the precision. K byeeeee.

🍓 list

Time to come back to perl reality from the PDL wonder land. list turns a pdl data object back into a regular perl list. 🎉

🫠 Caveat though: list produces a flat list. The dimensionality of the data object is lost. Gone. Cherish it while you had it.

🌈 Signature

This isn't a function, it's a concept. You'll see stuff like this all the time in the pod:

stdv

  Signature: (a(n); [o]b())

The signature tells you what the function wants as input and what kind of output it gives you. a(n) means it expects a 1D pdl with n elements. [o] is for output, and b() means it's a scalar. So stdv takes your 1D list and hands back a scalar. The name a or b or c is not important — what matters is the thing sitting in the parenthesis. 👑

corr

  Signature: (a(n); b(n); [o]c())

Here corr takes two inputs: two 1D pdls with the same number of elements. And it gives back one scalar.

t_test

  Signature: (a(n); b(m); [o]t(); [o]d())

Here t_test can take two 1D pdls of unequal size (n==m is certainly fine too, she's flexible), and gives back two scalars: the t-value and the degrees of freedom. Yes we accommodate t-tests with unequal sample sizes. We're thoughtful like that. 💖

assign

  Signature: (data(o,v); centroid(c,v); byte [o]cluster(o,c))

And here's one of the most complicated signatures in the whole package, from Kmeans. assign takes data of observation × variable dimensions, plus a centroid of cluster × variable dimensions, and returns an observation × cluster membership pdl (shown by 1s and 0s).

Got the idea? Then let's see how PDL does its magic :) ✨

🦋 Broadcasting

Another concept, bestie. And the first thing to know is: broadcasting is optional. 🫶

PDL broadcasting means automatically repeating the operation on extra elements or dimensions that you fed to a function. For a function with a signature like this:

gsl_cdf_tdist_P

  Signature: (double x(); double nu();  [o]out())

the signature says it takes two scalars as input and returns a scalar as output. So say you need to look up the p-values for a list of t's, all with the same degrees of freedom 19:

my @t = ( 1.65, 1.96, 2.56 );

my $p = gsl_cdf_tdist_P( pdl(@t), 19 );

print $p . "\n" . $p->info;

# here's what you will get

[0.94231136 0.96758551 0.99042586]
PDL: Double D [3]

The same function got repeated on each element in the list you provided. Ate. 💅

But what if you had different degrees of freedom for each t?

my @df = (199, 39, 19);

my $p = gsl_cdf_tdist_P( pdl(@t), pdl(@df) );

print $p . "\n" . $p->info;

# here's what you will get

[0.94973979 0.97141553 0.99042586]
PDL: Double D [3]

The df's get automatically matched with the t's to give you the results. Iconic behaviour, fr fr.

An example of broadcasting over extra dimension(s):

stdv

  Signature: (a(n); [o]b())

if the input is 2D — say you wanna compute the stdv for each of the 3 variables:

my @a = ( [1,1,3,4], [0,1,2,3], [4,5,6,7] );

# pdl @a is pdl dim [4,3]

my $sd = stdv( pdl @a );

print $sd . "\n" . $sd->info;

# this is what you will get

[ 1.2990381   1.118034   1.118034]
PDL: Double D [3]

So here the function got an input with an extra dimension of size 3, so it repeated the stdv operation on that extra dimension 3 times, and handed back a 1D pdl of size 3. Main character energy. 👑

Broadcasting works for an arbitrary number of dimensions — but bestie, please refrain from higher-dim pdls unless you've already decided to become a full PDL wiz / witch. 🙈

💀 Not all PDL::Stats methods broadcast. Rule of thumb: if a function has a signature attached to it, it broadcasts.

⭐ perldl

Essentially a perl shell with "use PDL;" at start up. It comes with the PDL installation, and it is very handy for trying out pdl operations, or just plain perl. print gets shortened to p so you don't injure yourself from excessive typing. 🥰

😭 Heads up: "my" goes out of scope at the end of (multi)line input, so mostly you're gonna have to drop the good practice of my here. Tragic.

🌸 For more info

PDL::Course 💜

---

👑 AUTHOR 👑

~~~~~~~~~~~~ ~~~~~ ~~~~~~~~ ~~~~~ ~~~ `` ><(((">

Copyright (C) 2009-2015 Maggie J. Xiong <maggiexyz users.sourceforge.net>

All rights reserved. There is no warranty. You are allowed to redistribute this software / documentation as described in the file COPYING in the PDL distribution.

---

And that's the whole page, bestie!! 💖 This one was rendered with perl v5.42.3 on 2026-08-21 — so if you're reading this way later, hi, hope your pdls are rectangular and your dimensions are tracked. 🎀✨🦋 Go do some stats. Periodt. 💅