Newsgroups: comp.ai.neural-nets,comp.ai.genetic,sci.op-research,sci.math
Path: cantaloupe.srv.cs.cmu.edu!das-news2.harvard.edu!news2.near.net!howland.reston.ans.net!ix.netcom.com!netcom.com!timin
From: timin@netcom.com (Mitchell E. Timin)
Subject: TSP CONTEST RESULTS (correction)
Message-ID: <timinCytuB2.7v0@netcom.com>
Summary: A correction to previously announced contest results
Keywords: TSP contest correction
Organization: NETCOM On-line Communication Services (408 261-4700 guest)
Date: Sun, 6 Nov 1994 03:42:37 GMT
Lines: 91
Xref: glinda.oz.cs.cmu.edu comp.ai.neural-nets:19906 comp.ai.genetic:4184 sci.op-research:2115 sci.math:85401



                    Results of the TSP Competition
                             (corrected)
OOPS!

In September I received a contest entry from Henning Klaskala of
Munich, Germany.  I later misplaced this program, and in late
October, when evaluating the various programs, this one was not with 
the others, and was totally overlooked.  After my first announcement,
last weekend, of the contest results Mr. Klaskala reminded me of his
program.  I then found it and proceeded to evaluate it by the same
procedures used for all of the others.  (These procedures will be
described in the near future.)  After doing so it was obvious that
Mr. Klaskala's entry was the best one!  Accordingly, he will be 
awarded a first place prize.  However the other contestants named
in the first announcement will still get the same prizes that they
would have gotten anyway.  This means that the total value of all
prizes will be increased by one first place prize.  (The specific
prizes awarded to each person will be announced in the near future.)
The new winners list follows:

This announcement concerns the TSP contest which was announced this
past spring and summer.  (For the curious reader who is not familiar
with this contest, e-mail to timin@netcom.com.)  A much more detailed
write-up will be posted within a week; this one is just to get the 
main results out quickly.

First Place in this Competition goes to Henning Klaskala of Germany!

Second Place goes to Jukka Honkanen of Finland!

Third Place goes to Frank Sven Nestel of Germany.

Fourth Place is for Randy Saint of Texas, U.S.A.

There is a three-way tie for fifth place between:
     Chad Hurwitz
     Juergen Klose
     The Koeln group

The above will receive a total of $1000 retail value of software,
plus one additional first place prize.  (Street Wizard mapping
software by Adept Computer Solutions)

A total of fifteen testable programs were received.  The programs
were each tested with three sets of data, for 47, 110 and 111 points.
They were each run several times to find out if their results were
randomly variable.  Mr. Klaskala's program was of this type; its
execution time was constant for a given problem file, but the routes
differed from run to run.  I ran his program seven times for each
input file and used the simple average of the route cost for
evaluation.  All of the other leading programs were deterministic,
giving the same route and execution time for every run.  Timing was
done with a stop watch, manually.  The same computer was used for all
runs; it is a 486DX66.

As examples of the performance of some of the better programs, Mr. 
Honkanen's program required only 20.3 seconds to find a route with
111 points, and the cost of that route was 88370.  Here is the route:

30 49 21 53 90 89 72 13 10 105 96 37 107 12 104 52 60 7 25 62 97 46
45 67 64 20 86 26 78 33 103 63 18 75 73 76 38 40 41 5 19 99 39 24 4
87 70 102 61 35 59 11 17 77 98 0 71 28 94 56 85 83 51 47 92 81 55 2
16 36 43 48 23 110 54 101 82 22 27 80 57 93 32 34 68 108 15 91 79 29
84 14 100 95 8 1 50 6 69 88 42 44 9 74 3 65 109 106 58 66 31 

Mr. Klaskala's program only required 1.5 seconds to find the
following route, with a cost of 88106:

30 104 52 60 7 62 25 37 107 12 21 49 53 90 89 72 13 79 91 29 34 32 15
108 68 84 14 100 95 8 1 69 6 50 58 44 42 88 93 57 80 27 106 22 65 109
71 28 74 3 101 82 54 110 92 47 51 85 83 94 56 9 102 61 35 59 11 77 0
98 17 87 70 39 24 4 99 19 5 41 38 40 76 73 81 55 2 16 36 18 75 43 48
63 103 20 86 26 33 78 64 67 23 45 46 97 66 105 96 10 31 

(The seven runs of the program had costs ranging from 87445 to 88480.)

The lowest cost route found for the 111 point case was done by Mr.
Klose's program in 83.3 seconds; its cost was 86800.  The route is:

30 104 10 13 79 90 89 72 53 49 21 52 60 7 37 107 12 25 62 97 46 45 67
64 20 86 26 78 33 103 63 18 75 36 16 55 2 92 81 73 76 38 40 41 51 47
85 83 94 56 9 102 61 35 50 6 69 8 1 95 100 14 59 11 17 87 70 39 24 4
99 19 5 0 98 77 71 28 74 3 65 109 44 42 88 58 57 93 32 34 91 29 84 68
108 15 80 27 106 22 101 82 54 110 43 48 23 66 105 96 31

The three problem files, for 47, 110 and 111 points, are available by
return e-mail, as is the initial contest announcement explaining how
to interpret these files.

