Author: Martin Schubert
Date: 14:56:08 06/11/01
Go up one level in this thread
On June 11, 2001 at 09:07:50, Steve Maughan wrote: >Hi Gian-Carlo, > >Yes the program uses the binomial distribution. It was a 'quick-n-dirty' app. >To expand on this; the program performs a hypothesis test that the two programs >are of equal strength i.e. would score 50% each in the long run. The binomial >distribution tells us that given the probability that anyone will win a single >game is 0.5 (i.e. equal strength), what the probability of winning 'x' games >from a total of 'y'. If this probability drops below the level of significance >then the long run probability is NOT 0.5 i.e. the player with the highest score >is actually better. > >Hope that explains it a bit better. Maybe I'll expand this in the next update. > >Regards, > >Steve Maughan Using the binomial distribution doesn't make much sense. The only sense is that you can be sure that a significant result according to the binomial distribution is really significant. It is not necessary to have the scores that you get with your program to get significant results. If you assume a probability for win,loose,draw of 1/3, just calculate the scores you need to get significant results for both models and compare them. You'll get a big difference. Regards, Martin
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.