查找列表中最长元素的最快方法(执行时间)

时间:2010-11-15 06:43:06

标签: perl algorithm performance

这是查找列表中最长元素的最快(执行时间)方式吗?

#!/usr/bin/env perl
use warnings;
use 5.012;
use List::Util qw(reduce);
use List::Util::XS;

my @array = qw( one two three four five six seven eight nine ten eleven );

my $l = reduce{ length($a) > length($b) ? $a : $b } @array;

say $l;

10 个答案:

答案 0 :(得分:17)

当只尝试查找列表中的一个元素时,不需要构建N大小的数据结构,因为这里有很多答案。执行此操作的最快O(N)方法是遍历阵列,跟踪最大元素。这样,您就可以O(N)访问列表,并O(1)使用内存。

sub longest {
    my $max = -1;
    my $max_i = 0;
    for (0 .. $#_) {              # for each index
        my $len = length $_[$_];  # only get length once per item
        if ($len > $max) {        # save index and update max if larger
            $max = $len;
            $max_i = $_;
        }
    }
    $_[$max_i]   # return the largest item
}

如果你要多次运行上面的代码,我会建议内联子程序的主体。

编辑:

drewk的基准测试显示,上面代码中的数组索引有点瓶颈。再尝试一下,我终于找到了一种比reduce解决方案更快的方法:

sub fastest {
    my $max = -1;
    my $max_ref;
    for (@_) {
        if (length > $max) {  # no temp variable, length() twice is faster
            $max = length;
            $max_ref = \$_;   # avoid any copying
        }
    }
    $$max_ref
}

导致以下基准:

           Rate longest   drewk  reduce fastest
longest 44245/s      --    -21%    -30%    -47%
drewk   55854/s     26%      --    -11%    -33%
reduce  63014/s     42%     13%      --    -25%
fastest 83638/s     89%     50%     33%      --

答案 1 :(得分:6)

以下是带有for和更少变量的OMG_peanuts的略微修改版本:

my $len = length $array[0];
my $longest = 0;
for my $i (1 .. $#array) {
    my $i_len = length $array[$i];
    if($i_len > $len) {
        $longest = $i;
        $len = $i_len;
    }
}
my $l = $array[$longest];

我正在使用基准测试,将其用于小数字(原始数组)

           Rate REDUCE TMPVAR TMPFOR
REDUCE 234862/s     --    -0%    -7%
TMPVAR 235643/s     0%     --    -6%
TMPFOR 251326/s     7%     7%     --

这适用于较大的数字或项目(原始数组x 100

         Rate TMPVAR TMPFOR REDUCE
TMPVAR 3242/s     --   -28%   -32%
TMPFOR 4503/s    39%     --    -5%
REDUCE 4750/s    47%     5%     --

请注意,算法的适用性因数据细节而有很大差异(我猜测更长的字符串可能会增加算法中length函数的权重)。

编辑:以下是基准测试的完整代码(长数组版本,数组定义中缺少x 100短消息)

use Benchmark  qw(:all);
use List::Util qw(reduce);

my @array = qw( one two three four five six seven eight nine ten eleven ) x 100;

cmpthese(-2, {
    REDUCE => sub {
        my $l = reduce{ length($a) gt length($b) ? $a : $b } @array;
    },
    TMPVAR => sub {
        my $idx = 1;
        my $lastLength = length $array[0];
        my $lastElt = $array[0];
        my $listLength = scalar @array;
        while ($idx < $listLength) {
          my $tmpLength = length $array[$idx];
          if ($tmpLength > $lastLength) {
            $lastElt = $array[$idx];
            $lastLength = $tmpLength
          }
          $idx++
        }
        my $l = $lastElt;
    },
    TMPFOR => sub {
        my $len = length $array[0];
        my $longest = 0;
        for my $i (1 .. $#array) {
            my $i_len = length $array[$i];
            if($i_len > $len) {
                $longest = $i;
                $len = $i_len;
            }
        }
        my $l = $array[$longest];
    },
});

答案 2 :(得分:3)

我最快的是:

sub drewk {
    my $len = -1;
    for (@_) {
        my $tmp=length($_);
        if ( $tmp > $len ) {
            $longest = $_;
            $len = $tmp;
        }
    }
    return $longest;
}

但反对:

sub strom {
    my $max = -1;
    my $max_i = 0;
    for (0 .. $#_) {              # for each index
        my $len = length $_[$_];  # only get length once per item
        if ($len > $max) {        # save index and update max if larger
            $max = $len;
            $max_i = $_;
        }
    }
    $_[$max_i]   # return the largest item
}

sub red {
    return reduce{ length($a) > length($b) ? $a : $b } @_;
}

显示reduce最快:

            Rate  strom  drewk reduce
strom  1323455/s     --   -38%   -45%
drewk  2144549/s    62%     --   -10%
reduce 2390707/s    81%    11%     --

另一个基准是Eric Strom's sub

答案 3 :(得分:2)

一条小金鱼:

my @unsorted = qw( one two three four five six seven eight nine ten eleven );
my $longest =  (
    map { $_->[0] } 
    sort { $b->[1] <=> $a->[1] } 
    map { [ $_, length $_ ] } @unsorted 
)[0];

say $longest;

编辑:对于熟悉该技术并想知道的人来说,地图/排序/地图是Schwartzian transform

答案 4 :(得分:2)

假设目标是找到最长的字符串,而不是其索引:

my $longest = $array[0];
my $len = length $longest;
for my $str (@array) {
    if ( length($str) > $len ) {
        $longest = $str;
        $len = length($str);
    }
}

答案 5 :(得分:1)

如果您真的想要截断计算出的length的数量,请查看Schwartian transform并将其用于您的问题。

编辑:

我看到没有人发布我的意思完整的例子,所以在这里(我还没有对它进行基准测试,因为我不是来自我的个人电脑):

my @array = qw( one two three four five six seven eight nine ten eleven );
my $longest = (
                reduce { $a->[1] > $b->[1] ? $a : $b } 
                map { [ $_, length $_ ] }
                @array
              )[0];

say $longest;

答案 6 :(得分:1)

这似乎比其他解决方案快得多(基于 faster_Eric_Storm ),

use warnings;
use 5.012;
use Benchmark qw(:all) ;
use List::Util qw(reduce);

my @array = map { ($_) x 50 } qw( one two three four five six seven eight nine ten eleven );

sub list_util_xs {
    my $l = reduce{ length($a) > length($b) ? $a : $b } @array;
    return $l;
}

sub fastest_Eric_Strom {
    my $max = -1; my $max_ref;
    for (@array) {
        if (length > $max) {
            $max = length;
            $max_ref = \$_;
        }
    }
    return $$max_ref;
}

sub ysth {
    my $longest = $array[0];
    my $len = length $longest;
    for my $str (@array) {
    if ( length($str) > $len ) {
        $longest = $str;
        $len = length($str);
    }
    }
    return $longest;
}

sub mpapec {
    my $max = -1;
    my $max_ref;
    length > $max and ($max, $max_ref) = (length, \$_) for @array;
    return $$max_ref;
}

cmpthese( -10, {
    'list_util_xs'  => sub{ list_util_xs() },
    'fastest_Eric_Storm' => sub{ fastest_Eric_Strom() },
    'ysth'      => sub{ ysth() },
    'mpapec'    => sub{ mpapec() },
});

输出

                      Rate list_util_xs fastest_Eric_Storm       ysth     mpapec
list_util_xs       13479/s           --               -24%       -24%       -29%
fastest_Eric_Storm 17662/s          31%                 --        -0%        -6%
ysth               17680/s          31%                 0%         --        -6%
mpapec             18885/s          40%                 7%         7%         --

答案 7 :(得分:0)

你可以使用一些temp var来避免一次又一次地计算长度:

my @unsorted = qw( one two three four five six seven eight nine ten eleven ); 
my $idx = 1;
my $lastLength = length $unsorted[0];
my $lastElt = $unsorted[0];
my $listLength = scalar @unsorted;
while ($idx < $listLength) {
  my $tmpLength = length $unsorted[$idx];
  if ($tmpLength > $lastLength) {
    $lastElt = $unsorted[$idx];
    $lastLength = $tmpLength
  }
  $idx++
}
print "Longest element:$lastElt";

基准测试结果:

           Rate REDUCE TMPVAR
REDUCE 169297/s     --   -29%
TMPVAR 237926/s    41%     --

答案 8 :(得分:0)

首先,我测试了所有例程是否给了我正确的结果。 MBO的例程没有通过第一个测试(它返回一个数组引用);给他第二个改变我修改了例程以获得正确的结果。

我有时会运行基准测试,但并不总是得到相同的顺序。 所以我会说(已经在这里发过)ysth和fasest_Eric_Strom是最快的但是list_utils reduce几乎和它们一样快; 从结果中可以很容易地看出,David Precious排序版本是最慢的,MBO修改后的缩减版本是第二慢的。

我的结论:list_utils reduce是最佳性价比的赢家 编辑: 颁奖仪式上我太快了: List::Util - reduce - length - encoding - question

David_Precious      64147/s             -- -36%        -73%               -79%  -80% -81%         -85%               -86% -87%
MBO                100195/s            56%   --        -58%               -67%  -69% -70%         -77%               -79% -80%
OMG_peanuts        237772/s           271% 137%          --               -21%  -27% -30%         -45%               -50% -52%
longest_Eric_Strom 300466/s           368% 200%         26%                 --   -8% -11%         -31%               -36% -40%
drewk              325883/s           408% 225%         37%                 8%    --  -4%         -25%               -31% -34%
bvr                338156/s           427% 237%         42%                13%    4%   --         -22%               -28% -32%
list_util_xs       434114/s           577% 333%         83%                44%   33%  28%           --                -8% -13%
fastest_Eric_Strom 471812/s           636% 371%         98%                57%   45%  40%           9%                 --  -5%
ysth               497198/s           675% 396%        109%                65%   53%  47%          15%                 5%   --

#!/usr/bin/env perl
use warnings;
use 5.012;
use Benchmark qw(:all) ;
use List::Util qw(reduce);

my @array = qw( one two three four five six seven eight nine very_long_long ten eleven );

sub list_util_xs {
    my $l = reduce{ length($a) > length($b) ? $a : $b } @array;
    return $l;
}

sub longest_Eric_Strom {
    my $max = -1; my $max_i = 0;
    for (0 .. $#array) {
        my $len = length $array[$_]; 
        if ($len > $max) { 
            $max = $len;
            $max_i = $_;
        }
    }
    return $array[$max_i];
}

sub fastest_Eric_Strom {
    my $max = -1; my $max_ref;
    for (@array) {
        if (length > $max) {
            $max = length;
            $max_ref = \$_;
        }
    }
    return $$max_ref;
}

sub David_Precious {
    my $longest =  ( map { $_->[0] } sort { $b->[1] <=> $a->[1] } map { [ $_, length $_ ] } @array )[0];
    return $longest;
}

sub MBO {
    my $longest = ( reduce { $a->[1] > $b->[1] ? $a : $b } map { [ $_, length $_ ] } @array )[0];
    return $longest->[0];
}

sub drewk {
    my $len = -1; my $longest;
    for (@array) {
        my $tmp=length($_);
        if ( $tmp > $len ) {
            $longest = $_;
            $len = $tmp;
        }
    }
    return $longest;
}

sub ysth {
    my $longest = $array[0];
    my $len = length $longest;
    for my $str (@array) {
    if ( length($str) > $len ) {
        $longest = $str;
        $len = length($str);
    }
    }
    return $longest;
}

sub bvr {
    my $len = length $array[0];
    my $longest = 0;
    for my $i (1 .. $#array) {
    my $i_len = length $array[$i];
    if($i_len > $len) {
        $longest = $i;
        $len = $i_len;
    }
    }
    return $array[$longest];
}

sub OMG_peanuts {
    my $idx = 1;
    my $lastLength = length $array[0];
    my $lastElt = $array[0];
    my $listLength = scalar @array;
    while ($idx < $listLength) {
    my $tmpLength = length $array[$idx];
    if ($tmpLength > $lastLength) {
    $lastElt = $array[$idx];
    $lastLength = $tmpLength
    }
    $idx++
    }
    return $lastElt;
}

cmpthese( -10, {
    'list_util_xs'  => sub{ list_util_xs() },
    'longest_Eric_Storm' => sub{ longest_Eric_Strom() },
    'fastest_Eric_Storm' => sub{ fastest_Eric_Strom() },
    'David_Precious'    => sub{ David_Precious() },
    'MBO'       => sub{ MBO() },
    'drewk'         => sub{ drewk() },
    'ysth'      => sub{ ysth() },
    'OMG_peanuts'   => sub{ OMG_peanuts() },
    'bvr'       => sub{ bvr() },
});

答案 9 :(得分:-1)

您可以通过缩减为包含字符串本身旁边长度的结构或数组来减少计算字符串长度的次数。

此外,通过reduce算法优化迭代({1}},length调用难以优化。