Reply
 
LinkBack Thread Tools Display Modes
  #1   Report Post  
Old July 17th 10, 10:00 PM posted to rec.games.chess.misc,rec.games.chess.computer
external usenet poster
 
First recorded activity by ChessBanter: Mar 2009
Posts: 3,536
Default How accurate are computer ratings compared to Fide Human ratings?Very accurate.

I estimate my playing strength based on how often I beat my PC chess
playing software (I have several programs, and a couple of handhelds),
whose Elo is calibrated according to the SSDF rating list found he
http://ssdf.bosjo.net/list.htm

The SSDF site has a interesting link to a Man Vs. Machine site:
http://privat.bahnhof.se/wb432434/Ma...2019%20Nov.htm

I often have wondered how accurate the SSDF computer rating list is,
that is, how close to a human Elo would a computer on the SSDF list
score, if the computer was a human, since, after all, the SSDF list
plays computers vs. computers with no human input. Turns out, it is
very accurate--an SSDF computer with a certain Elo corresponds very
close to a human with that Elo.

Proof is found in the Man vs. Machine link. Check out the matches by
Shreader 7, Shreader 9 and Rebel 10 (three that had long--and
therefore statistically significant--matches against humans). What I
found is as follows:

Shreader 9 (rated 2701 Elo according to SSDF) vs some human, Human
(rated 2381 Fide Elo). In a match of 20 games, computer Won 17, Drew
2 and Lost 1: 17-2-1, for a 90% margin. This means 374 point
difference. Hence the "predicted" human Elo for Shreader 9 is: 2381 +
374 = 2755 Fide Elo, vs the SSDF Elo of 2701. Pretty close (54 point
difference). More games would narrow this difference I'm sure.

Shreader 7 (rated 2680 Elo according to SSDF) vs Human (rated 2350
Fide Elo). Sco 14-6-0, or 85% margin, for a difference of 302
points. Predicted Elo: 2350 + 302 = 2652 Fide Elo, vs 2680 SSDF Elo.
Again, very close.

Rebel 10 (rated 2536 Elo according to SSDF) vs a human (rated 2393),
in a short match of 14 game, score was 7-5-2, or 68% margin, for a
difference of 137 points, or predicted Elo: 2393 + 137 = 2530 Fide
Elo, very close to the 2536 SSDF Elo.

Conclusion: The SSDF rating list corresponds closely to what a human,
if it was the computer, would score in a Fide tournament. Hence the
SSDF rating list is 'accurate'.

The only open question for me now, is what the relation between USCF
and Fide Elos are. I always assumed about from between 80 to 130
point difference (USCF being"inflated" relative to Fide Elo), but some
recent evidence (see he
http://groups.google.com/group/rec.g...8fd72891551f50
) suggests this gap has dropped. Hence my estimated USCF Elo may be
off a bit (I assumed I was about 1950 USCF Elo, with no time
controls*, but in fact I might be more like in the 1800s USCF Elo.

RL

* I play no time controls when playing my PC, meaning I have infinite
time to make a move. Hence this "inflates" my USCF Elo since I never
lose on time--in real life I would. But, in mitigation to this, the
machine also is set to "think on opponents time" so the more time I
take to make a move, the stronger the PC plays since it can figure out
more lines. It's hard to say what the trade off is.

One of these days I might have to enter a Fide tournament (Europe) or
USCF tourney (USA) and find out what my 'real' rating is, but this
requires I think about 25 or so games to be played, which is time
consuming.

  #2   Report Post  
Old July 17th 10, 10:10 PM posted to rec.games.chess.misc,rec.games.chess.computer
external usenet poster
 
First recorded activity by ChessBanter: Mar 2009
Posts: 3,536
Default How accurate are computer ratings compared to Fide Human ratings?Very accurate.

On Jul 18, 12:00*am, raylopez99 wrote:

The only open question for me now, is what the relation between USCF
and Fide Elos are. *I always assumed about from between 80 to 130
point difference (USCF being"inflated" relative to Fide Elo), but some
recent evidence (see hehttp://groups.google.com/group/rec.g...wse_thread/thr...
) suggests this gap has dropped. *Hence my estimated USCF Elo may be
off a bit (I assumed I was about 1950 USCF Elo, with no time
controls*, but in fact I might be more like in the 1800s USCF Elo.

RL

* I play no time controls when playing my PC, meaning I have infinite
time to make a move. *Hence this "inflates" my USCF Elo since I never
lose on time--in real life I would. *But, in mitigation to this, the
machine also is set to "think on opponents time" so the more time I
take to make a move, the stronger the PC plays since it can figure out
more lines. It's hard to say what the trade off is.

One of these days I might have to enter a Fide tournament (Europe) or
USCF tourney (USA) and find out what my 'real' rating is, but this
requires I think about 25 or so games to be played, which is time
consuming.


Of interest more to me than anybody reading this thread, I just reran
the numbers and I was off by 45 points. So I'm rated in the high
1880s to low 1900s USCF Elo I estimate, not mid 1900s. This is
because there's another rule for estimating microprocessor playing
strength when you halve the playing time on your chess program. Since
the SSDF rating list is based on 180 seconds per move (3 minutes), but
in fact I usually set it at 45 seconds to 23 seconds a move, you have
to estimate the drop in playing strength when you do this. The
estimates (with "think on opponents time") are between 75 to 140
points--I usually say 100 points.

So essentially, at the low Class A, high Class B level, I'm probably
slightly stronger than TK's last USCF Elo rating from a generation ago
(TK retired from competitive chess about 20 years ago it seems, and
was around in the mid 1800s USCF Elo), weaker than MikeMurray, Sam
Sloan, who play at the high Class A level, and weaker than SBD, who is
playing at the Expert level. Sounds about right.

The wildcard for me, however, is the "infinite time to move" factor.
Like I say, I have unlimited time--akin to a postal game--to think
about a move when playing my PC. I'm not sure how much this would
affect by OTB play, but I'm sure I would drop some points because of
it in a real Fide or USCF tournament.

RL
  #3   Report Post  
Old July 18th 10, 01:03 AM posted to rec.games.chess.misc,rec.games.chess.computer
external usenet poster
 
First recorded activity by ChessBanter: Oct 2007
Posts: 3,073
Default How accurate are computer ratings compared to Fide Human ratings?Very accurate.

On Jul 17, 5:00*pm, raylopez99 wrote:

One of these days I might have to enter a Fide tournament (Europe) or
USCF tourney (USA) and find out what my 'real' rating is, but this
requires I think about 25 or so games to be played, which is time
consuming.


Don't bother, from your games you'reobviously almost an IM.


  #4   Report Post  
Old July 18th 10, 01:31 AM posted to rec.games.chess.misc,rec.games.chess.computer
external usenet poster
 
First recorded activity by ChessBanter: Dec 2009
Posts: 1,146
Default How accurate are computer ratings compared to Fide Human ratings?Very accurate.

On Jul 17, 5:00*pm, raylopez99 wrote:

The only open question for me now, is what the relation between USCF
and Fide Elos are. *I always assumed



In fact the differences between USCF and FIDE ratings have varied
considerably over time, so your policy of 'always assuming' things is
a bit of a reckless approach. A superior method is to always NOT
assume anything. That way the worst that can happen is you will
occasionally miss an opportunity to make a 'safe' assumption.


about from between 80 to 130
point difference (USCF being"inflated" relative to Fide Elo), but some
recent evidence (see hehttp://groups.google.com/group/rec.g...wse_thread/thr...
) suggests this gap has dropped. *Hence my estimated USCF Elo may be
off a bit (I assumed I was about 1950 USCF Elo, with no time
controls*, but in fact I might be more like in the 1800s USCF Elo.



You now seem to be equating USCF ratings to 'Elo' -- quite a flip
around even for you, Phil.


* I play no time controls when playing my PC, meaning I have infinite
time to make a move. *



Wrong. If you take *that long* to consider a move, you will
invariably die. Much better then to stick with finite periods
which fall well within the normal lifespan of humans.


Hence this "inflates" my USCF Elo since I never
lose on time--in real life I would. *But, in mitigation to this, the
machine also is set to "think on opponents time" so the more time I
take to make a move, the stronger the PC plays since it can figure out
more lines. It's hard to say what the trade off is.



Not hard at all! The computer quickly becomes obsolete,
while you -- much more slowly -- age, deteriorate and finally,
turn back to the dust from whence you came. One key
difference is that while the computer is essentially a slave to
humans you of course are a free agent, able to fritter your
whole life away here in rgc-- if that is what you wish to do. Of
course, only a certified nutcase would do that.


One of these days I might have to enter a Fide tournament (Europe) or
USCF tourney (USA) and find out what my 'real' rating is, but this
requires I think about 25 or so games to be played, which is time
consuming.



In fact you can get a 'provisional' USCF rating by playing in
a single rated event, perhaps completed in but a single day.
This happens all the time in our local events. A new player
shows up, pays his USCF dues and entry fee, plays say, five
rounds against the regulars, losing most of them while at the
same time struggling to remember to press his clock and
accurately write down the moves, and then goes home dead
broke having finished in last place. If he happens to return the
following month and is lucky enough to win or draw a few
games, the pairings program sends him to me-- for another
round of arse-whooping mayhem! Then (for some reason or
other) we never see the poor fellow again.

Maybe the *real* reason you don't want to enter an OTB
tournament is that deep down you know you will get killed.
This is why you continue to spew lame excuses -- it beats
the alternative: OTB massacre!



  #5   Report Post  
Old July 18th 10, 02:42 AM posted to rec.games.chess.misc,rec.games.chess.computer
sd sd is offline
external usenet poster
 
First recorded activity by ChessBanter: Jun 2009
Posts: 922
Default How accurate are computer ratings compared to Fide Human ratings?Very accurate.

On Jul 17, 7:26*pm, OldHaasie wrote:
My top USCF rating was 2044. *My last published is1954 dating back to
1989 or thereabout. *However, I always felt like I played no better
than a 1600 USCF player throughout my entire amateur career.

Old Haasie


But you didn't have to spend all your time conjecturing what your
rating might be based on blitz games played against faceless opponents
on the Internet, or "take ten points for ...b5," you actually went out
there and fought and did better than most. Perhaps your ability to be
critical of your own play was helpful in that regard.

I think you should be proud of the fact that you were in there
swinging and did better than you thought you could.

SBD

Reply
Thread Tools
Display Modes

Posting Rules

Smilies are On
[IMG] code is On
HTML code is Off
Trackbacks are On
Pingbacks are On
Refbacks are On


Similar Threads
Thread Thread Starter Forum Replies Last Post
Chess Ratings samsloan alt.chess (Alternative Chess Group) 6 May 10th 08 07:33 AM
An old article about early chess computers! Quite amusing rant. [email protected] rec.games.chess.computer (Computer Chess) 0 January 21st 07 06:32 AM
Should the USCF rate the Olympiads? Sam Sloan rec.games.chess.politics (Chess Politics) 9 March 29th 06 11:52 PM
Should the USCF rate the Olympiads? Sam Sloan rec.games.chess.misc (Chess General) 9 March 29th 06 11:52 PM
Should the USCF rate the Olympiads? Sam Sloan alt.chess (Alternative Chess Group) 9 March 29th 06 11:52 PM


All times are GMT +1. The time now is 08:42 PM.

Powered by vBulletin® Copyright ©2000 - 2019, Jelsoft Enterprises Ltd.
Copyright 2004-2019 ChessBanter.
The comments are property of their posters.
 

About Us

"It's about Chess"

 

Copyright © 2017