Author: Martin Schubert
Date: 23:52:37 05/25/02
Go up one level in this thread
On May 25, 2002 at 17:46:11, Peter Fendrich wrote:
>On May 25, 2002 at 17:19:19, Rolf Tueschen wrote:
>
>>On May 25, 2002 at 16:00:01, Peter Fendrich wrote:
>>
>>>You and some other guys keep saying this ("Nothing") but that is not a good
>>>interpretation of the list.
>>>The ratings are the best predictions of each engines strength. Fritz is still
>>>the best prediction of who is the best even if it's a quite unsafe prediction.
>>
>>This is absolutely false. And at the first glance it's both true and false.
>>
>>- it's true because now, at the moment of the cut of the information stream
>>FRITZ was 8 points above the second (but the moment of the cut is important;
>>without further information we cannot judge whether the cut was good or bad)
>>
>>- it's false, because (and this is trivial) with 8 points advance and a margin
>>of error of 30 points all could happen in future; either FRITZ on place 1 or
>>place two, even place 5. It is absolutely false that the first place at the
>>deliberate moment of the cut has any predictive power more than for place two at
>>the moment of the cut and a future of place 1 or 5.
>>
>>- it's absolutely false overall, because we have _no_ information about the
>>future. Therefore Sandro Necchi and all critics of SSDF are right. And nobody
>>even didn't start to talk about different hardware, different samples of
>>opponents, and the validation of the data with human chessplayers for the
>>meaning of the Elo numbers.
>>
>>
>>>The current error margin just says tell us that we can't be 95% sure. Lower the
>>>expectations of probability and the error margin intervals will shrink.
>>>
>>>Peter
>>
>>THe presentation of the SSDF ranking list tells us, that although the SSDF
>>defense is always hinting at the no-science argument, but still the list is made
>>to inspire the fantasy in the clients of a scientific project because of the
>>sophisticated margins and probabilities. The critic however discovers that SSDF
>>does not obey the simplest rule of experiments namely the control of the
>>variables and the holding them constant, to be able to get as a result the Elo
>>numbers of the rating. If all is flexible, you'll never know what your results
>>should stand for. That you and the SSDF has no bad feelings is simply a result
>>of your own expectations. As long as the results "look" like normal you think
>>that your test design must be ok. But chess testing it's also known that a
>>result can look ok for the wrong reasons.
>>
>>Rolf Tueschen
>
>Sigh... We had this discussion some years ago. You didn't know what you talked
>about then. Just long nonsense posts. You still doesn't know what you are
>talking about. Your arguments aren't developed one inch since then.
>Peter
I don't know if he knows what he is talking about. But the things he critizices
are right. The SSDF doesn't think about experimental setups, they just throw
results into a formula and get a result. That has not much to do with statistics
IMO.
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.