Nerl.pm

filter: Statementstmt 47.2% 26 / 55 Branchbran 5.0% 1 / 20 Conditioncond 0.0% 0 / 12 MC/DCmcdc 0.0% 0 / 8 Subroutinesub 55.5% 5 / 9 Podpod 0.0% 0 / 6 totaltotal 29.0% 32 / 110 CC 15 SCAR 41.8
CC
15
Cov
31%
init_network
132.0
append_l2
6.0
train
6.0
CRAP 65.1
↑ Summary Nerl/Network.pm »
1 package AI::Nerl;
2 2 ▶use Moose 'has',inner => { -as => 'moose_inner' };
Subroutine BEGIN: called CC=1 SCAR=0.0
3 2 ▶use PDL;
Subroutine BEGIN: called CC=1 SCAR=0.0
4 2 ▶use AI::Nerl::Network;
Subroutine BEGIN: called CC=1 SCAR=0.0
5  
6 # ABSTRACT: Neural networks with backpropagation.
7  
8 # main_module
9  
10 2 our $VERSION = .03;
11  
12 #A Nerl is a mechanism to build neural networks?
13 #Give it training,test, and cv data?
14 #it settles on a learning rate and stuff?
15 #or maybe it's also a language for guided training?
16 #or maybe a visual gui thing?
17 #Not exactly sure. Maybe I'm tinkering with forces better left alone.
18 #That's a great excuse for failing horribly.
19  
20  
21 =head1 AI::Nerl - A sort of stackable neural network builder thing.
22  
23 =head1 SYNOPSIS
24  
25 Check out L<AI::Nerl::Network>; This module is in an early stage.
26  
27 =head1 AUTHOR
28  
29 Zach Morgan
30  
31 =cut
32  
33  
34  
35 2 has scale_input => (
36 is => 'ro',
37 isa => 'Num',
38 required => 0,
39 default => 0,
40 );
41 2 has l2 => ( #hidden layer.
42 is => 'ro',
43 isa => 'Num',
44 default => 30,
45 );
46  
47 2 has [qw/ train_x
48 train_y /] => (
49 is => 'ro',
50 isa => 'PDL',
51 required => 0, #training can be done manually.
52 );
53 2 has [qw/ test_x cv_x
54 test_y cv_y /] => (
55 is => 'ro',
56 isa => 'PDL',
57 required => 0,
58 );
59  
60 2 has network => (
61 required=>0,
62 is => 'rw',
63 isa => 'AI::Nerl::Network',
64 );
65  
66 2 has passes=> (
67 is => 'rw',
68 isa => 'Int',
69 default => 10,
70 );
71  
72 2 has basis => (
73 is => 'ro',
74 isa => 'AI::Nerl',
75 required => 0,
76 );
77  
78 #initialize $self->network, but don't train.
79 # any parameters AI::Nerl::Network takes are fine here.
80 0 ▶sub init_network{
Subroutine init_network: not called CC=11 SCAR=48.8
Pod undocumented
81 0 my $self = shift;
82 0 my %nn_params = @_;
83 #input layer size:
84 0 ▶ unless ($nn_params{l1}){
Branch: 1 branch
BranchTrueFalseTotal
unless ($nn_params{"l1"}) 0 0 0
85 0 ▶ if ($self->basis){
Branch: 2 branches
BranchTrueFalseTotal
if ($self->basis) { } 0 0 0
elsif ($self->train_x) { } 0 0 0
86 0 $nn_params{l1} = $self->basis->network->l1 + $self->basis->network->l2;
87 } elsif($self->train_x) {
88 0 ▶ $nn_params{l1} ||= $self->train_x->dim(1);
Condition0%0/3
$nn_params{"l1"} ||= $self->train_x->dim(1)
l!l&&r!l&&!r
000
MC/DC0%0/2
$nn_params{"l1"} ||= $self->train_x->dim(1)
$nn_params{"l1"} $self->train_x->dim(1)
Truth table0%0/3
A ||= B
A= $nn_params{"l1"}B= $self->train_x->dim(1)
ABresult
1X1
011
000
89 }
90 }
91 #output layer size:
92 0 ▶ unless ($nn_params{l3}){
Branch: 1 branch
BranchTrueFalseTotal
unless ($nn_params{"l3"}) 0 0 0
93 0 ▶ if ($self->basis){
Branch: 2 branches
BranchTrueFalseTotal
if ($self->basis) { } 0 0 0
elsif ($self->train_x) { } 0 0 0
94 0 $nn_params{l3} = $self->basis->network->l3;
95 } elsif($self->train_x) {
96 0 ▶ $nn_params{l3} ||= $self->train_y->dim(1);
Condition0%0/3
$nn_params{"l3"} ||= $self->train_y->dim(1)
l!l&&r!l&&!r
000
MC/DC0%0/2
$nn_params{"l3"} ||= $self->train_y->dim(1)
$nn_params{"l3"} $self->train_y->dim(1)
Truth table0%0/3
A ||= B
A= $nn_params{"l3"}B= $self->train_y->dim(1)
ABresult
1X1
011
000
97 }
98 }
99 0 ▶ $nn_params{l2} ||= $self->l2;
Condition0%0/3
$nn_params{"l2"} ||= $self->l2
l!l&&r!l&&!r
000
MC/DC0%0/2
$nn_params{"l2"} ||= $self->l2
$nn_params{"l2"} $self->l2
Truth table0%0/3
A ||= B
A= $nn_params{"l2"}B= $self->l2
ABresult
1X1
011
000
100 0 ▶ $nn_params{scale_input} ||= $self->scale_input;
Condition0%0/3
$nn_params{"scale_input"} ||= $self->scale_input
l!l&&r!l&&!r
000
MC/DC0%0/2
$nn_params{"scale_input"} ||= $self->scale_input
$nn_params{"scale_input"} $self->scale_input
Truth table0%0/3
A ||= B
A= $nn_params{"scale_input"}B= $self->scale_input
ABresult
1X1
011
000
101  
102 0 my $nn = AI::Nerl::Network->new(
103 %nn_params
104 );
105 0 $self->network($nn);
106 }
107  
108 2 ▶sub build_network{
Subroutine build_network: called CC=1 SCAR=0.0
Pod undocumented
109 2 my $self = shift;
110 2 my $nn = AI::Nerl::Network->new(
111 l1 => $self->train_x->dim(1),
112 l2 => $self->l2,
113 l3 => $self->train_y->dim(1),
114 scale_input => $self->scale_input,
115 );
116 2 $nn->train($self->train_x, $self->train_y, passes=>$self->passes);
117 2 $self->network($nn);
118 }
119  
120 0 ▶sub append_l2{
Subroutine append_l2: not called CC=2 SCAR=17.9
Pod undocumented
121 0 my ($self,$x) = @_;
122 0 ▶ if($self->basis){
Branch: 1 branch
BranchTrueFalseTotal
if ($self->basis) 0 0 0
123 0 $x = $self->basis->append_l2($x);
124 }
125 0 return $self->network->append_l2($x);
126 }
127  
128  
129 1 ▶sub run{
Subroutine run: called CC=2 SCAR=7.4
Pod undocumented
130 1 my ($self,$x) = @_;
131 1 $x->sever;
132 1 ▶ if($self->basis){
Branch: 1 branch
BranchTrueFalseTotal
if ($self->basis) 0 1 1
133 0 $x = $self->basis->append_l2($x);
134 }
135 1 return $self->network->run($x);
136 }
137 0 ▶sub train{
Subroutine train: not called CC=2 SCAR=17.9
Pod undocumented
138 0 my ($self,$x,$y) = @_;
139 0 $x->sever;
140 0 ▶ if($self->basis){
Branch: 1 branch
BranchTrueFalseTotal
if ($self->basis) 0 0 0
141 0 $x = $self->basis->append_l2($x);
142 }
143 0 return $self->network->train($x,$y);
144 }
145  
146 0 ▶sub cost{
Subroutine cost: not called CC=2 SCAR=17.9
Pod undocumented
147 0 my ($self,$x,$y) = @_;
148 0 $x->sever();
149 0 ▶ if($self->basis){
Branch: 1 branch
BranchTrueFalseTotal
if ($self->basis) 0 0 0
150 0 $x = $self->basis->append_l2($x);
151 }
152 0 return $self->network->cost($x,$y);
153 }
154  
155 2 'a neural network has your dog.';
↑ Summary Nerl/Network.pm »