Author: Albert Silver
Date: 16:48:24 09/24/00
Go up one level in this thread
On September 24, 2000 at 07:31:48, Jason Williamson wrote:
>On September 24, 2000 at 00:17:48, Albert Silver wrote:
>
>>On September 24, 2000 at 00:08:55, Jason Williamson wrote:
>>
>>>On September 23, 2000 at 19:55:13, Albert Silver wrote:
>>>
>>>>On September 23, 2000 at 12:16:51, Jason Williamson wrote:
>>>>
>>>>>On September 22, 2000 at 17:36:32, Stephen Ham wrote:
>>>>>
>>>>>>Dear Terry,
>>>>>>
>>>>>>Jeroen is correct. One should use database statistics only as a very general
>>>>>>guide.
>>>>>>
>>>>>>Consider this: Somebody plays a tricky new move and wins with it. Others
>>>>>>discover the tricky new move and win with it too. Let's assume that after a
>>>>>>period of time, the line acumulates 9 wins. However, at some point, an opponent
>>>>>>finally busts the line because it was not intrinsically sound. So the fact is,
>>>>>>the line is no good, but the statistics say that out of 10 games played, it won
>>>>>>90% of the time. So Terry, what are you going to believe, the statistics or the
>>>>>>truth?
>>>>>>
>>>>>>Database statistics would thus lead one to belive that good lines are "bad" and
>>>>>>bad lines are "good". Instead, my advice, Terry, is to forget about what that
>>>>>>statistics say and play openings that you both think are good/best and you feel
>>>>>>comfortable playing.
>>>>>>
>>>>>>Good luck, Terry.
>>>>>>
>>>>>>Stephen
>>>>>
>>>>>The above doesn't really hold, since if enough peopple (read 9) think its good
>>>>>for the tricky move and lose with it, the database will suddenly be at 50%. So
>>>>>eventually, the refuted move will correct it self.
>>>>
>>>>Not unless the players don't keep their databases up-to-date.
>>>>
>>>> Albert Silver
>>>>
>>>>>This of course assumes you
>>>>>update your database.
>>>>>
>>>>>On the other hand, nothing will match your own eye, and a good opening book on
>>>>>the variations you want to study.
>>>>>
>>>>>
>>>>>JW
>>>
>>>Hmm I think I said that one line below where you put your answer.
>>>
>>>JW
>>
>>I thought maybe you hadn't noticed it which is why I repeated it. If the players
>>update their databases and see the refutation, why on earth should the stats
>>eventually balance out?
>>
>> Albert Silver
>
>Quite simply, because not everyone has databases, thus weaker players will still
>play this promising looking move, these games would there way in to databases,
>and then result will be that it would even out.
>
>If you look in a large database, almost every line is in there, except perhaps
>the most obvious blunders (and even there, its surprising how many are in the
>databases).
>
>JW
It all depends on the database then. Although I have two large databases full of
junk, they are not the ones I use for any non player-specific research. In other
words, I only preserve them for research on players, bad or good. For opening
research I use databases where games of players rated below 2000 (when possible)
were filtered out. The larger commercial databases (such as CB's Mega2000 or
CA's Hugebase) usually come fairly filtered anyhow so that you are unlikely to
ever see stats of weaker players balancing out. In any case it also means that a
careful look at the statictical material presented will be necessary as it will
present even more misleading results.
There was a discussion some time back in RGCC, in which KK began a new trend in
his 'tips of the week', posting 'winning' statistics. He started by posting that
1.d4 was the way to go because the stats 'didn't lie', and went on to talk about
winning chances by keeping queens on the board in certain openings. Again,
thanks to statistics. I remember agreeing and suggested 1.a4 was even better,
and then followed this up by a bunch of ridiculous (but true)
counter-statistics. Bob (Hyatt) mentioned that he had run into this when Crafty
began playing Nf3-Ng1 as an opening sequence because of its new learning
function based only on results. The GMs were not amused.
Albert Silver
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.