--- In [email protected], "meldeiry" > > You guys are all trying to explain why DICrOTIC sims 6 points better > than DIChOTIC, but the fact is that it doesn't. It only does that > much better in the championship player report that I ran. When you > sim them head to head, they're within a point of each other.
One important distinction: when you choose a move you do not have to simulate every move to the same number of iterations, which you would have to do if you want to compare two moves. For example, if you do a few hundred iterations and find that DICrOtic has results that are 6 points higher than DIChOTIC, then you can safely conclude that there are more important issues to spend your time on. You can devise statistically sound versions of the reasoning above, which distribute simulation attention to the move that makes that greatest contribution to error rate. Such systems can reduce error rates as fast as possible using whatever time is available. It is quite common to see highly skewed iteration counts. Maven will typically simulate only two moves to the maximum iteration count, and might simulate hundreds of moves to a small number of iterations. This is a natural outcome of the distribution of quality of moves in Scrabble; usually only a handful of moves are close to the best. I don't know if Quackle does the same. Brian
