Search found 4782 matches

by Laskos
Tue Jul 28, 2009 10:35 pm
Forum: Computer Chess Club: General Topics
Topic: ELO rating inflation
Replies: 9
Views: 1250

Re: ELO rating inflation

A stronger computer getting a lower rating would indicate rating deflation rather than rating inflation. Incorrect, an even stronger computer getting a higher rating, would indicate inflation. Depends on your perspective -:) Eh? An "even stronger computer" would be expected to get a higher rating, ...
by Laskos
Tue Jul 28, 2009 12:58 pm
Forum: Computer Chess Club: General Topics
Topic: ELO rating inflation
Replies: 9
Views: 1250

Re: ELO rating inflation

I did enquire a few years back with Hyatt why his Crafty program ELO had decreased over time rather than increased, which one would expect with better hardware and software updates. Hyatt's response was that more computer opponents play on ICC, which have caused a decrease in ELO, i.e. higher ELO o...
by Laskos
Mon Jul 27, 2009 1:23 pm
Forum: Computer Chess Club: General Topics
Topic: ELO rating inflation
Replies: 9
Views: 1250

Re: ELO rating inflation

I wonder if this article would also apply to the SSDF and computer chess in general? http://chessbase.com/newsdetail.asp?newsid=5608 Cordially, Sean :) Sonas is probably right. I also investigated the problem, and came to the conclusion that the players reaching the rating floor can deliver inflati...
by Laskos
Wed Jul 22, 2009 7:59 am
Forum: Computer Chess Club: General Topics
Topic: Diminishing returns of increasing search depth
Replies: 43
Views: 8459

Re: Diminishing returns of increasing search depth

Don't scare me, I made a bet with Eelco on a bottle of a good Champagne for the 1st of August 2017. I did two Pade fits, each with 5 parameters, one with asymptotic behaviour 1/x, second with 1, throwing away the first 3 results (the data are from ply 4 up). The data of the first (1/x) fit up to ply...
by Laskos
Tue Jul 21, 2009 8:49 am
Forum: Computer Chess Club: General Topics
Topic: Diminishing returns of increasing search depth
Replies: 43
Views: 8459

Re: Diminishing returns of increasing search depth

I calculated the aggregate elo at each depth, arbitrarily setting the strength at 1 ply to 0. The result certainly looks logarithmic to me, although this is hardly a rigorous test. Fit Curve, A*log(B*x+D)+C (A=3846.48, B=0.349316, C=-603.192, D=1.11704) rms = 26.8582, chi-square = 2.81317, sigma = ...
by Laskos
Tue Jul 21, 2009 12:30 am
Forum: Computer Chess Club: General Topics
Topic: Diminishing returns of increasing search depth
Replies: 43
Views: 8459

Re: Diminishing returns of increasing search depth

ply ELO difference 2/1 416 3/2 326 4/3 224 5/4 152 6/5 224 7/6 168 8/7 132 9/8 168 10/9 116 11/10 135 12/11 135 13/12 116 14/13 105 A relation that seems to fit the data is: log(e^3 / d^2) / 4.5 = 1 where: e is ELO difference d is depth difference in % As example for 3/2 326 e = 326 d = ...
by Laskos
Mon Jul 20, 2009 12:53 am
Forum: Computer Chess Club: General Topics
Topic: Diminishing returns of increasing search depth
Replies: 43
Views: 8459

Re: Diminishing returns of increasing search depth

It is not a linear scale, the x-axis is more a logarithmic scale. What kind of statistics do you need for that? I do not know. Eelco A small correction, plies are logarithmic with time, Elo is also logarithmic, so Elo-plies is linear (assumed constant nearly 70 for two-fold increase in time). It is...
by Laskos
Sun Jul 19, 2009 2:32 pm
Forum: Computer Chess Club: General Topics
Topic: Diminishing returns of increasing search depth
Replies: 43
Views: 8459

Re: Diminishing returns of increasing search depth

Maybe Bob can help us with Crafty to get a solution? His testing cluster could give us thousands of games and more ply depth in test results. It would be interesting to compare my data to data that Crafty produces. - Jarkko I think you can do it, to 1000 games, and it is worth doing it. Below 10-12...
by Laskos
Sun Jul 19, 2009 2:23 am
Forum: Computer Chess Club: General Topics
Topic: Diminishing returns of increasing search depth
Replies: 43
Views: 8459

Re: Diminishing returns of increasing search depth

This is of course very nice, a correlation of 0.990988 but why do you think you can extrapolate these data almost ad infinitum? I would not trust any figure beyond 16 plies if you have experimental data running only till 14 plies. It is not a linear scale, the x-axis is more a logarithmic scale. Wh...
by Laskos
Sun Jul 19, 2009 12:53 am
Forum: Computer Chess Club: General Topics
Topic: Diminishing returns of increasing search depth
Replies: 43
Views: 8459

Re: Diminishing returns of increasing search depth

Toga II 1.4beta5c played against itself. Search depth difference was fixed so that other searched one ply less. Results games in one test: 102 Error bounds 95% confidence about ±55 ELO ply ELO difference 2/1 416 3/2 326 4/3 224 5/4 152 6/5 224 7/6 168 8/7 132 9/8 168 10/9 116 11/10 135 12/11 135 13...