RedDevils - The Strategy

Basic elements of our strategy

• 1. Always throws 8192...12288 in very first throw
• 4. Never throws 1...127 in very first throw
• 10. Strategy does not change at all from sequence to sequence
• 14. Adapts to partners throws immediately

Skeleton of our strategy

The following is a rough description of our strategy. It is mainly based on the ideas presented in the Scientific American, May/June 1986, (sorry, forgot the titles... see section "mathematical games").

• Try to cooperate by winning/losing alternately (Be cooperative!)
• If cooperation like this doesnt work: (Be provocative!)

  Handle five different cases: (assume our last guess was p, 'theirs' q)

  1. q > 2*p

  This is good for us, if p is large. Issue (q-1)/2 as next value if p is large enough, or 2*q else to gain as many points as possible.

  2. 2*p >= q > p

  This situation is very bad. Back down to 2 as soon as possible.

  3. p = q

  We can't do much here. To avoid deadlocks, do a non-deterministic (i.e. random) throw.

  4. p > q >= p/2

  Once again, this is good. Issue 2*q as next value to gain as many points as possible.

  5. p/2 > q

  This again is a bad situation. Try to throw a bigger number next time (that is, 2*q).

• Try to go back to cooperation mode now and then (Be forgiving!)

Improvements in step 2

• count points, and if non-cooperation is better, don't cooperate.
• In the final throw we do a 'surprise throw', since we have nothing to lose... We still need a good random function that yields the maximum Erwartungswert for our last throw.
• If we are losing heavily, perhaps we should go to throwing 1s only.

Room for for even more improvements

• Our method of 'exploring' enemies is very simple. Although it proves to be effective against many simple strategies, it can also be explored quite easily. The next version will probably be less deterministic in making a throw, especially in the cases 4., 5. and also 1. (see above).
• Almost all additional ideas that have not been mentioned in the above 'Step 2' description were usesless or even bad when included. Overall the code could be greatly simplified by removing these obsolete parts.

Wettbewerbe
Meine Homepage
UNIX-AG Homepage