This is a slight modification of Don Reble s idea on that page , I simply replaced Pi by a strategy, which is much harder to outlearn ... at least for my strategies:
class BlindPig extends Strategy
{
private int pi[] = {6,3,3,2,2,2,6,3,1,1};
public int initial_turn()
{
return pi[0];
}
public int turn(int last_turns[], int scores[],
int carryOver, int turn_no)
{
return pi[turn_no % pi.length];
}
}
Simply look, what happens:
Hmm, so much about outlearning a simple peridic Strategy. Obviously all the initial gains by TimeWarp just base on the other loosers, which drop out even faster. As soon as all but NFocusPQ and TimeWarp are dead, even they die away...
Ideas I had in BlindPig:
Keep it period 10, to keep it similar ...
Use a threefold repetition of digits, since TimeWarp cannot spot it.
Use also repetition on a higher pattern level (6-3).
Use lower numbers, like those on the Theory Page .
T he other thing would be to set up an experimental environment, to find the optimal period-N (e.g.N=10) strategy according to a given environment ... But this sounds like crying for a very, very fast computer. Not just a mere PC ...