Author: Andrew Dados
Date: 06:44:52 12/31/01
Suppose I am getting tons of scores for some experiment which outcome will obey known distribution (In my problem it is Poisson distribution; type of distribution should not matter). I can't store all scores, but I need to know average and mean parameters, so I could recreate distribution function at some time later. How can I store some set of data as small as possible to be able to add new scores to it and still get my mean/sigma right? Example: One experiment is 1000 tosses of a coin. In this case outcome is number of heads. I will collect unspecified number of such results. In this case I could simply store an array of 1000 counters, but I can't afford it. Average number can be easily stored and incrementally updated with 2 ints: total sum and number of experiments. Can some similar trick be done to recalculate mean value after new score comes in? Chess example (closer to my problem): I have a chess position for which I am getting time-to-solve results from many players. So their rating distribution is 'predefined' here. The more samples I will collect, the more accurately I can assing a rating for some new player solving this position. I can not collect all separate times-to-solve. So for each player I need to update some totals to be able to calculate mean from those totals (average is easy). Can this be accurately done? ..and no... while it sounds like that - it is not some school assignment. :) -Andrew-
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.