Sections
Home
Hills
Infinite Hills
Tournaments
Software
Evolving
Optimizer
Community
Newsletter
Discussion
History
Sections
 
For Beginners
First Steps
FAQ
Guides
Lexicon
Benchmarks
For Beginners
> Home > The Corewar Lexicon > '94 Paper > Miscellous Hints

6 Miscellous Hints:


6.1 Paper and Quickscanner:

If you're working on a successful paper, usually you will test/benchmark your idea against scanners and oneshots to be sure the constants are well chosen. If you have finalized your paper, don't sit back and relax, because that's not all that can be done!!!

This chapter focuses on the use of an additional quick-scanner beside the paper. "Well, I know" one could say, "nobody would write a paper without a quick-scanner". But nothing is published so far about the influence in scoring if it is used.

Some of the advantages are evident. The paper will gain some more wins even if it battles against another paper (which usually ends close to 100% ties). Also, keep in mind modern quick-scanning warriors win about 10-15% of the battle (if fighting against another quick-scanning warrior) only with their quick-scanning part.

But how is the situation against a scanner? Usually they don't use a quick-scanner. One effect should be that the quickscanner part acts additionally as a decoy which should be good against scanner. But how strong is the influence in reality? To answer this question, I decided to have a closer look, focusing on a variety of different papers and scanners.

The aspect we want to figure out first is: How much influence a decoy has on the scoring of a paper against scanner. Usually they should score slightly better, because the scanner will be entraped to wipe the "useless" decoy and give the paper a bit more time for a lucky hit or to gain enough processes to survive.

The test warriors looks like:
;code of the paper is here (copy'n'pasted from the original source. 
                            Also includes the original bootstrapping)

Decoy equ length

for (85-Decoy)
dat 0, 0
rof

i for Decoy
dat <i, >-i
rof

end
The following scanner were used for the test: Claw, Herbal Avenger, myBlur2, Stalker, Win!, Willow, Zoom, Origin of Storm.

And here are the scoring for eight different papers, having various length decoys:
BR: Benji's Revenge;  D: Disincentive;  F: Fixed;  nP: nPaper II;
RP: Paper of Reepicheep;  RoP: Revenge of the Papers
MJ: Mini Return of the Jedimp;  RF: Return of Fugitive

Decoy
length    BR      D       F       nP     RP      RoP     MJ      RF
---------------------------------------------------------------------
0       126,0   141,6   148,9   133,5   152,6   120,9   123,1   146,3
10      125,4   142,9   146,9   134,8   152,0   123,6   125,6   147,9
20      123,4   143,5   148,8   131,3   152,1   123,1   124,6   148,9
30      126,2   145,3   149,7   132,2   153,4   123,9   125,7   149,3
40      128,3   147,9   149,8   135,9   151,9   126,4   124,7   152,5
50      126,3   146,1   150,7   138,0   152,8   125,8   124,7   152,4
60      123,9   145,2   149,1   136,6   151,0   126,2   124,7   153,1
70      128,8   143,1   148,5   137,6   153,7   126,4   126,5   154,1
80      128,1   143,2   149,8   138,9   151,8   127,2   128,0   154,1

highest
increase  2,8     6,3     1,8     5,4     1,1     6,3     4,9     7,8
The results weren't very surprising. All paper gain a few points by having a decoy. Interesting is the inconsistency of the points increase depending on the length of the decoy for the different papers. If we look to the average scores for the length of the decoy we can see more clearly that the optimal decoy should have a length of at least 30 instructions. The Q^3 should fit this length while the mini Q^4 is a bit too small to act sucessfully as a decoy.
length Average  incr.
---------------------
 0      136,6   0,0
10      137,4   0,8
20      137,0   0,4    \-> mini Q^4
30      138,2   1,6    /
40      139,7   3,1    --> Q^3
50      139,6   3,0
60      138,7   2,1    --> Q^2
70      139,8   3,2
80      140,1   3,5
Now let's have a look how well the quick-scanners alone scores against the scanners. Both quick-scanners start at the end of the scanning phase a djn 0, #200. The quick-scanner will die after 200 additional cycles to be sure all kills due the bombing are recorded. The results shows as expected, that the Q^3 scores much better than the mini Q^4.
qscanner    length    pts.     +decoy
-------------------------------------
  Q^3         42      15,6      18,7
 mQ^4         26      10,4      12,0
If we now use both quick-scanners together with the paper we get the following results:
length    BR      D       F       nP     RP      RoP     MJ      RF
---------------------------------------------------------------------
Q^3     136,3   153,5   158,8   142,0   161,1   140,7   136,3   162,2
mQ^4    129,4   148,1   156,6   135,4   156,8   136,9   129,9   154,6
The points increase compared to the pure paper are shown below:
Q^3     10,3    11,9    9,9     8,5     8,5     19,8    13,1    15,9
mQ^4    3,4     6,5     7,6     1,9     4,1     16,1    6,8     8,4
Well, that's what we've expected after the tests with the quick-scanner. The reason why the points increase isn't as high as expected from the tests above can be explained with the delayed launching of the paper, because the quick-scanner will executed first. Interesting is the behaviour of Revenge of the Paper, because it scores with both quick-scanners much better than expected. The reason seem to be that it only launches the two copies of itself and don't use the code in front of the quick-scanner.

Finally we can say that a Q^3 quickscanner is the better choice if the opponent is a scanner. Against other strategies it can respond differently. But this will be discussed as well as the influence of booting later.
2002-2005 corewar.info. Logo C. Schmidt
Paper

'88 Paper

'94 Paper
Description
The Basics
The Offense
Paper with Imps
Uncommon Papers
How to Optimize a Paper
Miscellaneous Hints

LP Paper

Paper