Offshore
Worker Accident
Counter
Uli's
cEvo Homepage
Contact:
ulrich.olderdissen@gmx.de
Last
update
any page: Nov. 4, 2009
Last
update
this page: Nov. 4, 2009
General
remarks
- The
AI is written in MS
C++ version 8. I will send you the source
code if you ask for it.
- Please help with the debug
by
sending me (zipped) saved files of any bugs, even if they are not
repeatable. Now that Steffen
has supplied a reload function I can find even random bugs with a good
chance. I promise
to
fix the bug and
will send you a fixed AI as soon as possible.
Software Quality
Before
release I do at least
50 bug free games in tournament
mode at the stated design point(s). So at the design point
the AI
should be relatively bug free. Since the server
does not
support tournaments at random senarios, I
have no
practical way of testing those and one is to expect some bugs
in
scenarios off the design point.
For release 0.1475 testing at additional scenarios (see table on the
release page) has been done, so that overall at least 1000 error free
games have been played.
Misnomer
Actually
the AI's in this game have no artificial
intelligence whatsoever. They are simply a collection of
behavioral rules that act on a given input with a
given
output.
If there should be no rule for an input, at best they do nothing, but
most likely they throw up their bits and abend.
A
true AI
requires learning from past experience, that is, the AI needs
feedback. However, in our case there is no feedback from the
field at
all. Also, learning is only possible if the same game is
played a
multiple of times so that the AI can do some play variations and
thereby find out the best play. But
this is done
only at the design level. So the AI reflects
only the learning and experience of the designer.
Even
the
most modern chess programs are no AI's in the true sense, they
simply calculate the best move by brute compute power. Now
this
game is much more complex. Just consider the number of chess
pieces (32), decreasing during the game, against the number of units
there might be in cEvo.
Release Numbering
The
release
numbering follows the release number of cEvo. So for the cEvo
release 0.14.7 AI_UO will also have the release number
0.14.7.
Any
released fixes would then be numbered 0.14.7.n.
However, due to the fact that some game rules have been
changed drastically, in particular the behavior of the
settlers, this AI will not be updated to cEvo
release 1.xx.
Design Point/s
Release
0.147.3:
Presently,
there
is just one design point: 100% map, 90 % land,
seven
opponents.
At other scenarios,
the AI should work,
but
definitely not at its best.
Release
0.147.4_sea:
100% map, 30% land,
one
to seven
opponents
Release 0.147.5:
230% map, 30% land, one opponent
Critique
All
critique is
welcome. However, critique is meaningless if the scenario you
are
critizising is not close to the stated design point/s.
That
is like
complaining that your PORSCHE handles badly under water. You
should get a submarine.
The
same holds for complaints about the compute speed of this AI.
Only if there is an AI that can beat AI_UO at least half of the time at
the design point and also is faster, such critique would be
founded. Remember: function costs compute time.
Releases/Downloads
Some remarks on
history and
experience.
Some
proposals on how to improve the server-AI interface.
Table
of best developers.