Author: Dann Corbit
Date: 18:33:33 06/03/04
Go up one level in this thread
On June 03, 2004 at 20:30:29, Peter Fendrich wrote: >On June 03, 2004 at 12:46:39, Dan Wulff wrote: > >>Hi Peter! >> >>On June 02, 2004 at 19:04:15, Peter Fendrich wrote: >> >>>forget about performance rating (and your statement about 16 points is wrong...) >> >>Wrong, how ??? The formula for Expected score is: >> >>1/(1+(diff/400)) which will always give a number less than 1. >> >>to calculate the win/loss you need (Score-Expected score)*16 for master players >>(*32 for lower ratings). How should that ever be more than 16 points, master >>games ?? >> >>Greetings >> >>Dan Wulff >>(The Gandalf Team) > >Hello Dan, >Yes I agree, you're right, but the values 16 and 32 are selected more or less by >random... >I don't think that performance rating is the right way to go. It is not intended >to be used for one game only. >IMO it's clearly better to use the expected score based on the rating diff. > >For instance if the rating diff is 100 the expected score is 0.64 for the higer >rated player. I can send you a table with expected scores depending on the >rating difference if you want to. I think that you could use that figure instead >of rating points. > >Now, with the expected score 0.64 you can translate it to an ELO >increment/decrement that makes the ELO difference to be kept to 100 when a lot >of games are played. 64% of n (many) games should keep the ratings unchanged for >these two players. >In the table (from the link I gave you) you will find that the higher rated >player earns 12 points per winning game. Suppose that we have a 100 game match >with the expected result 64-36. 64*12 - 36*20 should be 0 but it's not. That's >strange, I thought that the LASK system should fit! Well, if you want to >translate to ELO points it's probably better to adjust the table to fit your own >purpose. It's not a very hard task with the expected score. Yes, I think this is the much better approach. Imagine a die with only the numbers 1,2,3 on it twice, rolled against a regular dice that has 1 through 6 on it. If we get a 3 with the strange die and a 2 with the regular die, what are we to conclude? We know that (on an average basis) the regular die underperformed on this roll, but we need a larger collection to make statistical inferences (e.g. is the die with 1-6 shaved or weighted...). On the other hand, suppose that we have some chess position, and three moves are routinely made from that position. Suppose (further) that each is played at least 30 times in the database. Now, we can look at the performance of each move against the expected Elo and see if each move underperforms, overperforms, or performs as expected on average.
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.