line |
stmt |
bran |
cond |
sub |
pod |
time |
code |
1
|
|
|
|
|
|
|
package Statistics::Sampler::Multinomial::AliasMethod;
|
2
|
|
|
|
|
|
|
|
3
|
1
|
|
|
1
|
|
93622
|
use 5.014;
|
|
1
|
|
|
|
|
3
|
|
4
|
1
|
|
|
1
|
|
3
|
use warnings;
|
|
1
|
|
|
|
|
1
|
|
|
1
|
|
|
|
|
24
|
|
5
|
1
|
|
|
1
|
|
3
|
use strict;
|
|
1
|
|
|
|
|
1
|
|
|
1
|
|
|
|
|
31
|
|
6
|
|
|
|
|
|
|
|
7
|
|
|
|
|
|
|
our $VERSION = '0.7';
|
8
|
|
|
|
|
|
|
|
9
|
1
|
|
|
1
|
|
3
|
use Carp;
|
|
1
|
|
|
|
|
1
|
|
|
1
|
|
|
|
|
64
|
|
10
|
1
|
|
|
1
|
|
504
|
use Ref::Util qw /is_arrayref/;
|
|
1
|
|
|
|
|
428
|
|
|
1
|
|
|
|
|
64
|
|
11
|
1
|
|
|
1
|
|
5
|
use List::Util qw /min sum/;
|
|
1
|
|
|
|
|
1
|
|
|
1
|
|
|
|
|
72
|
|
12
|
1
|
|
|
1
|
|
471
|
use List::MoreUtils qw /first_index/;
|
|
1
|
|
|
|
|
6939
|
|
|
1
|
|
|
|
|
4
|
|
13
|
1
|
|
|
1
|
|
456
|
use Scalar::Util qw /blessed/;
|
|
1
|
|
|
|
|
2
|
|
|
1
|
|
|
|
|
43
|
|
14
|
|
|
|
|
|
|
|
15
|
1
|
|
|
1
|
|
443
|
use parent qw /Statistics::Sampler::Multinomial/;
|
|
1
|
|
|
|
|
205
|
|
|
1
|
|
|
|
|
6
|
|
16
|
|
|
|
|
|
|
|
17
|
|
|
|
|
|
|
sub _validate_prng_object {
|
18
|
8
|
|
|
8
|
|
13
|
my ($self, $prng) = @_;
|
19
|
|
|
|
|
|
|
|
20
|
|
|
|
|
|
|
# Math::Random::MT::Auto has boolean op overloading
|
21
|
|
|
|
|
|
|
# so make sure we don't trigger it or our tests fail
|
22
|
|
|
|
|
|
|
# i.e. don't use "if $prng" directly
|
23
|
|
|
|
|
|
|
# (and we waste a random number, but that's less of an issue)
|
24
|
8
|
100
|
|
|
|
17
|
return 1 if !defined $prng;
|
25
|
|
|
|
|
|
|
|
26
|
7
|
50
|
|
|
|
24
|
croak 'prng arg is not an object'
|
27
|
|
|
|
|
|
|
if not blessed $prng;
|
28
|
7
|
50
|
|
|
|
26
|
croak 'prng arg does not have rand() method'
|
29
|
|
|
|
|
|
|
if not $prng->can('rand');
|
30
|
|
|
|
|
|
|
|
31
|
7
|
|
|
|
|
82
|
return 1;
|
32
|
|
|
|
|
|
|
}
|
33
|
|
|
|
|
|
|
|
34
|
|
|
|
|
|
|
sub _initialise_alias_tables {
|
35
|
8
|
|
|
8
|
|
22
|
my ($self, %args) = @_;
|
36
|
|
|
|
|
|
|
|
37
|
8
|
|
|
|
|
11
|
my $probs = $self->{data};
|
38
|
|
|
|
|
|
|
|
39
|
|
|
|
|
|
|
# caller has not promised they sum to 1
|
40
|
8
|
50
|
|
|
|
16
|
if (!$self->{data_sum_to_one}) {
|
41
|
8
|
|
|
|
|
27
|
my $sum = sum (@$probs);
|
42
|
8
|
50
|
|
|
|
22
|
if ($sum != 1) {
|
43
|
8
|
|
|
|
|
12
|
my @scaled_probs = map {$_ / $sum} @$probs;
|
|
111
|
|
|
|
|
99
|
|
44
|
8
|
|
|
|
|
17
|
$probs = \@scaled_probs;
|
45
|
|
|
|
|
|
|
}
|
46
|
|
|
|
|
|
|
}
|
47
|
|
|
|
|
|
|
|
48
|
|
|
|
|
|
|
# algorithm and comments stolen/adapted from
|
49
|
|
|
|
|
|
|
# https://hips.seas.harvard.edu/blog/2013/03/03/the-alias-method-efficient-sampling-with-many-discrete-outcomes/
|
50
|
|
|
|
|
|
|
|
51
|
8
|
|
|
|
|
8
|
my (@smaller, @larger);
|
52
|
8
|
|
|
|
|
23
|
my @J = (0) x scalar @$probs;
|
53
|
8
|
|
|
|
|
19
|
my @q = (0) x scalar @$probs;
|
54
|
8
|
|
|
|
|
8
|
my $kk = -1;
|
55
|
8
|
|
|
|
|
10
|
my $K = scalar @$probs;
|
56
|
|
|
|
|
|
|
|
57
|
8
|
|
|
|
|
12
|
foreach my $prob (@$probs){
|
58
|
111
|
|
|
|
|
67
|
$kk++;
|
59
|
111
|
|
|
|
|
86
|
$q[$kk] = $K * $prob;
|
60
|
111
|
100
|
|
|
|
116
|
if ($q[$kk] < 1.0) {
|
61
|
65
|
|
|
|
|
54
|
push @smaller, $kk
|
62
|
|
|
|
|
|
|
}
|
63
|
|
|
|
|
|
|
else {
|
64
|
46
|
|
|
|
|
37
|
push @larger, $kk;
|
65
|
|
|
|
|
|
|
}
|
66
|
|
|
|
|
|
|
}
|
67
|
|
|
|
|
|
|
|
68
|
|
|
|
|
|
|
# Loop though and create little binary mixtures that
|
69
|
|
|
|
|
|
|
# appropriately allocate the larger outcomes over the
|
70
|
|
|
|
|
|
|
# overall uniform mixture.
|
71
|
8
|
|
100
|
|
|
29
|
while (scalar @smaller && scalar @larger) {
|
72
|
101
|
|
|
|
|
75
|
my $small = pop @smaller;
|
73
|
101
|
|
|
|
|
55
|
my $large = pop @larger;
|
74
|
|
|
|
|
|
|
|
75
|
101
|
|
|
|
|
68
|
$J[$small] = $large;
|
76
|
101
|
|
|
|
|
85
|
$q[$large] = ($q[$large] + $q[$small]) - 1;
|
77
|
|
|
|
|
|
|
|
78
|
101
|
100
|
|
|
|
101
|
if ($q[$large] < 1.0) {
|
79
|
37
|
|
|
|
|
65
|
push @smaller, $large;
|
80
|
|
|
|
|
|
|
}
|
81
|
|
|
|
|
|
|
else {
|
82
|
64
|
|
|
|
|
116
|
push @larger, $large;
|
83
|
|
|
|
|
|
|
}
|
84
|
|
|
|
|
|
|
}
|
85
|
|
|
|
|
|
|
# handle numeric stability issues
|
86
|
|
|
|
|
|
|
# courtesy http://www.keithschwarz.com/darts-dice-coins/
|
87
|
8
|
|
|
|
|
18
|
while (scalar @larger) {
|
88
|
9
|
|
|
|
|
8
|
my $g = shift @larger;
|
89
|
9
|
|
|
|
|
19
|
$q[$g] = 1;
|
90
|
|
|
|
|
|
|
}
|
91
|
8
|
|
|
|
|
15
|
while (scalar @smaller) {
|
92
|
1
|
|
|
|
|
4
|
my $l = shift @smaller;
|
93
|
1
|
|
|
|
|
3
|
$q[$l] = 1;
|
94
|
|
|
|
|
|
|
}
|
95
|
|
|
|
|
|
|
|
96
|
|
|
|
|
|
|
# need better names for these,
|
97
|
8
|
|
|
|
|
12
|
$self->{J} = \@J;
|
98
|
8
|
|
|
|
|
10
|
$self->{q} = \@q;
|
99
|
|
|
|
|
|
|
|
100
|
8
|
100
|
|
|
|
48
|
return if !defined wantarray;
|
101
|
|
|
|
|
|
|
|
102
|
|
|
|
|
|
|
# should not expose these to the caller
|
103
|
3
|
|
|
|
|
9
|
my %results = (J => \@J, q => \@q);
|
104
|
3
|
50
|
|
|
|
20
|
return wantarray ? %results : \%results;
|
105
|
|
|
|
|
|
|
}
|
106
|
|
|
|
|
|
|
|
107
|
|
|
|
|
|
|
sub draw {
|
108
|
10
|
|
|
10
|
1
|
7310
|
my ($self, $args) = @_;
|
109
|
|
|
|
|
|
|
|
110
|
10
|
|
|
|
|
18
|
my $prng = $self->{prng};
|
111
|
|
|
|
|
|
|
|
112
|
|
|
|
|
|
|
my $q = $self->{q}
|
113
|
10
|
|
66
|
|
|
26
|
// do {$self->_initialise_alias_tables; $self->{q}};
|
|
4
|
|
|
|
|
11
|
|
|
4
|
|
|
|
|
8
|
|
114
|
|
|
|
|
|
|
|
115
|
10
|
|
|
|
|
10
|
my $J = $self->{J};
|
116
|
10
|
|
|
|
|
12
|
my $K = scalar @$J;
|
117
|
10
|
|
|
|
|
27
|
my $kk = int ($prng->rand * $K);
|
118
|
|
|
|
|
|
|
|
119
|
|
|
|
|
|
|
# Draw from the binary mixture, either keeping the
|
120
|
|
|
|
|
|
|
# small one, or choosing the associated larger one.
|
121
|
10
|
100
|
|
|
|
30
|
return ($prng->rand < $q->[$kk]) ? $kk : $J->[$kk];
|
122
|
|
|
|
|
|
|
}
|
123
|
|
|
|
|
|
|
|
124
|
|
|
|
|
|
|
sub draw_n_samples {
|
125
|
4
|
|
|
4
|
1
|
6561
|
my ($self, $n) = @_;
|
126
|
|
|
|
|
|
|
|
127
|
4
|
|
|
|
|
9
|
my $prng = $self->{prng};
|
128
|
|
|
|
|
|
|
|
129
|
|
|
|
|
|
|
my $q = $self->{q}
|
130
|
4
|
|
66
|
|
|
14
|
// do {$self->_initialise_alias_tables; $self->{q}};
|
|
1
|
|
|
|
|
4
|
|
|
1
|
|
|
|
|
3
|
|
131
|
4
|
|
|
|
|
9
|
my $J = $self->{J};
|
132
|
4
|
|
|
|
|
7
|
my $K = scalar @$J;
|
133
|
|
|
|
|
|
|
|
134
|
4
|
|
|
|
|
12
|
my @draws = (0) x $K;
|
135
|
4
|
|
|
|
|
13
|
for (1..$n) {
|
136
|
41
|
|
|
|
|
51
|
my $kk = int ($prng->rand * $K);
|
137
|
|
|
|
|
|
|
# Draw from the binary mixture, either keeping the
|
138
|
|
|
|
|
|
|
# small one, or choosing the associated larger one.
|
139
|
|
|
|
|
|
|
# {SWL: could try to use Data::Alias or refaliasing here
|
140
|
|
|
|
|
|
|
# as the derefs cause overhead, albeit the big overhead
|
141
|
|
|
|
|
|
|
# is the method calls}
|
142
|
|
|
|
|
|
|
#push @draws, ($prng->rand < $q->[$kk]) ? $kk : $J->[$kk];
|
143
|
41
|
100
|
|
|
|
72
|
$draws[($prng->rand < $q->[$kk]) ? $kk : $J->[$kk]]++;
|
144
|
|
|
|
|
|
|
}
|
145
|
|
|
|
|
|
|
|
146
|
4
|
|
|
|
|
9
|
return \@draws;
|
147
|
|
|
|
|
|
|
}
|
148
|
|
|
|
|
|
|
|
149
|
|
|
|
|
|
|
# Cuckoo package to act as a method wrapper
|
150
|
|
|
|
|
|
|
# to use the perl PRNG stream by default.
|
151
|
|
|
|
|
|
|
package Statistics::Sampler::Multinomial::AliasMethod::DefaultPRNG {
|
152
|
|
|
|
|
|
|
sub new {
|
153
|
1
|
|
|
1
|
|
9
|
return bless {}, __PACKAGE__;
|
154
|
|
|
|
|
|
|
}
|
155
|
|
|
|
|
|
|
sub rand {
|
156
|
12
|
|
|
12
|
|
126
|
rand();
|
157
|
|
|
|
|
|
|
}
|
158
|
|
|
|
|
|
|
1;
|
159
|
|
|
|
|
|
|
}
|
160
|
|
|
|
|
|
|
|
161
|
|
|
|
|
|
|
1;
|
162
|
|
|
|
|
|
|
__END__
|