NAME
Peta::NN::Data - records with named fields, to train models on and measure them by
VERSION
version 0.2610090
SYNOPSIS
use Peta::NN::Data;
my $nouns = Peta::NN::Data->read('nouns.tsv', fields => [qw(singular gender plural listed)])
->derive(length => sub ($noun) { length $noun->{singular} })
->hold_out(0.1)
->mark(core => sub ($noun) { length $noun->{listed} });
printf "%d nouns, %d of them held out, %d the core\n",
$nouns->count, $nouns->held->count, $nouns->marked('core')->count;
my $pairs = $nouns->shown->pairs(from => 'singular', to => 'plural', given => ['gender']);DESCRIPTION
A record is one thing a model is to learn something about, as a table of named fields. A model is told which field it reads, which it answers and which it is given beside, and takes its training pairs from the data.
Whether a record is held out of training follows from a field's value, not from chance: the same record is on the same side on every run and every machine, and records that share the value stay together. A record can also be marked as belonging to a named group, such as a core a model has to get right without exception.
where, shown, held, marked and sample give views of the same records; nothing is copied.
METHODS
new
Peta::NN::Data->new(records => [ { ... }, ... ], fields => [ names ]).
read
Peta::NN::Data->read($file, fields => [ names ]): one record per tab-separated line; lines that start with # are comments.
records
The records, as a list of tables.
count
How many records there are.
fields
The names of the fields.
values_of
values_of($field): that field's value of every record.
derive
derive(name => sub ($record) { ... }): a new field, worked out from each record. Returns the data.
where
where(sub ($record) { ... }): a view of the records the sub accepts.
hold_out
hold_out($share, by => $field, never => $mark): holds that share of the records out of training, decided by the value of a field (the first one unless named, or what a sub gives for a record). The records of the mark never names are not held out. Returns the data. Data made with new without fields has its fields in alphabetical order, so name the field there.
hold_out_if
hold_out_if(sub ($record) { ... }): holds out the records the sub accepts, and no others. Returns the data.
shown
A view of the records that are trained on.
held
A view of the records that are held out.
mark
mark(name => sub ($record) { ... }): names the group of records the sub accepts. Returns the data.
marks
The names of the groups.
marked
marked($name): a view of that group.
is_marked
is_marked($name, $record): whether the record is in that group.
is_held
is_held($record): whether the record is held out.
sample
sample($count): a view of at most that many records, the same ones every time, spread evenly over the whole.
pairs
pairs(from => $field, to => $field, given => [ fields ]): for each record [input, answer, parameters ...], as a model is trained on.
AUTHOR
PetaMem s.r.o. <info@petamem.com>
COPYRIGHT
Copyright (c) 2026 PetaMem s.r.o.
LICENSE
This package is free software, dual-licensed under the Artistic License 2.0 and the BSD 2-Clause License. See the LICENSE file of the distribution.