in reply to Re: Random data generation.
in thread Random data generation.

A working solution, but like others that invoke the regex engine, it proves slower, especially for longer strings, than a flags & conditions solution. (Eg. salva's 846681). 

[16:16:52.15] c:\test>junk -I=3 ABCDEF 12
      Rate   x0   x3   x8   x4   x6   x5   x7   
x0  9284/s   --  -7% -16% -33% -38% -38% -47%
x3  9946/s   7%   -- -10% -29% -33% -34% -44%   ikegami
x8 11091/s  19%  12%   -- -20% -26% -26% -37%   johngg
x4 13951/s  50%  40%  26%   --  -7%  -7% -21%   ikegami variant
x6 14952/s  61%  50%  35%   7%   --  -0% -15%
x5 14964/s  61%  50%  35%   7%   0%   -- -15%
x7 17647/s  90%  77%  59%  26%  18%  18%   --   salva

[16:17:13.48] c:\test>junk -I=3 ABCDEF 120
     Rate   x0   x3   x4   x8   x6   x5   x7
x0  279/s   --  -8% -10% -78% -83% -83% -91%
x3  302/s   8%   --  -3% -76% -81% -81% -90%
x4  310/s  11%   3%   -- -75% -81% -81% -89%
x8 1250/s 348% 314% 303%   -- -22% -22% -58%
x6 1595/s 472% 428% 414%  28%   --  -0% -46%
x5 1597/s 472% 429% 415%  28%   0%   -- -46%
x7 2951/s 958% 877% 851% 136%  85%  85%   --

[16:17:30.86] c:\test>junk -I=3 ABCDEF 1200
     Rate    x0    x4    x3    x8    x6    x5    x7
x0 3.38/s    --   -1%   -3%  -97%  -98%  -98%  -99%
x4 3.42/s    1%    --   -1%  -97%  -98%  -98%  -99%
x3 3.46/s    3%    1%    --  -97%  -98%  -98%  -99%
x8  121/s 3478% 3433% 3388%    --  -24%  -24%  -63%
x6  159/s 4608% 4548% 4489%   32%    --   -1%  -51%
x5  160/s 4633% 4573% 4513%   32%    1%    --  -51%
x7  324/s 9499% 9377% 9256%  168%  104%  103%    --

[download]

As you can see. salva's solution wins easily on performance.

Replies are listed 'Best First'.
Re^3: Random data generation.
by almut (Canon) on Jun 27, 2010 at 18:27 UTC

    I wonder if you've overlooked my solution, or if there's anything wrong with it.

    Not only is it simpler (both less lines of code and conceptually easier to understand, IMHO), it's also consistently faster than what you found to be the fastest: 

    $ ./846784.pl
        Rate salva almut
salva  881/s    --  -15%
almut 1037/s   18%    --

    [download]

    (tested with v5.10.1, x86_64-linux-thread-multi)

    Read more... benchmark (1315 Bytes)
[reply]
[d/l]
[select]
      I wonder if you've overlooked my solution, or if there's anything wrong with it.

      Indeed I did, as it was posted as a reply to someone else. And no, there is nothing wrong with it.

      It beats out salva's (just), and is a simple, clean, and direct implementation of the requirements. I have my winner. Thank you.

      Now I've seen it, I'm somewhat chagrined that I didn't see it for myself. It would still be my winner even if it were slightly slower than the flags & resets versions.

      Yours is xA below: 

      [19:48:48.65] c:\test>junk -I=3 ABCDEF 12
      Rate   x0   x3   x9   x8   x5   x4   x6   x7   xA
x0  9794/s   --  -8% -13% -19% -35% -36% -38% -43% -51%
x3 10606/s   8%   --  -6% -12% -30% -31% -33% -39% -46%
x9 11286/s  15%   6%   --  -6% -25% -26% -29% -35% -43%
x8 12063/s  23%  14%   7%   -- -20% -21% -24% -30% -39%
x5 15122/s  54%  43%  34%  25%   --  -1%  -5% -12% -24%
x4 15340/s  57%  45%  36%  27%   1%   --  -3% -11% -23%
x6 15863/s  62%  50%  41%  32%   5%   3%   --  -8% -20%
x7 17248/s  76%  63%  53%  43%  14%  12%   9%   -- -13%
xA 19822/s 102%  87%  76%  64%  31%  29%  25%  15%   --

[19:49:50.63] c:\test>junk -I=3 ABCDEF 120
     Rate    x0    x3    x4    x8    x9    x6    x5    x7    xA
x0  275/s    --  -20%  -21%  -80%  -80%  -86%  -86%  -91%  -92%
x3  341/s   24%    --   -2%  -75%  -75%  -82%  -83%  -89%  -90%
x4  349/s   27%    2%    --  -75%  -75%  -82%  -82%  -89%  -90%
x8 1383/s  404%  305%  296%    --   -0%  -29%  -30%  -57%  -59%
x9 1386/s  405%  306%  297%    0%    --  -29%  -30%  -57%  -59%
x6 1945/s  608%  470%  457%   41%   40%    --   -1%  -40%  -42%
x5 1971/s  618%  477%  465%   43%   42%    1%    --  -39%  -41%
x7 3216/s 1071%  842%  821%  133%  132%   65%   63%    --   -4%
xA 3346/s 1119%  880%  859%  142%  141%   72%   70%    4%    --

[19:50:43.49] c:\test>junk -I=3 ABCDEF 1200
     Rate     x4     x0     x3     x8     x9     x6     x5     x7     
+xA
x4 3.33/s     --    -0%    -8%   -98%   -98%   -98%   -98%   -99%   -9
+9%
x0 3.33/s     0%     --    -8%   -98%   -98%   -98%   -98%   -99%   -9
+9%
x3 3.61/s     8%     8%     --   -97%   -98%   -98%   -98%   -99%   -9
+9%
x8  137/s  4019%  4019%  3700%     --    -8%   -25%   -27%   -61%   -6
+3%
x9  149/s  4366%  4366%  4020%     8%     --   -19%   -21%   -58%   -6
+0%
x6  183/s  5392%  5392%  4966%    33%    23%     --    -3%   -48%   -5
+0%
x5  189/s  5578%  5578%  5138%    38%    27%     3%     --   -46%   -4
+9%
x7  351/s 10434% 10434%  9618%   156%   136%    92%    86%     --    -
+5%
xA  368/s 10960% 10960% 10104%   169%   148%   101%    95%     5%     
+--

      [download]
[reply]
[d/l]
        The length of the generated data is a parameter that does not modify the ranking because the costs of all the given solutions are linear to it.

        If you change the set size or the maximun number of consecutive repetitions instead, then the ranking is not so stable. The benchmarks below show it.

        I have also added a new solution salva2 (much faster in average) and some variations of almut's one.

        The code:

        Read more... (3 kB)

        And the benchmarking results:

        Read more... (9 kB)
        Note that the salva2 algorithm has some bias. It is able to generate all the valid solutions but not all of them have the same probability.
[reply]
[d/l]
[select]

In Section Seekers of Perl Wisdom