Author: James B. Shearer
Date: 13:54:41 10/28/99
Go up one level in this thread
On October 28, 1999 at 06:27:10, Amir Ban wrote:
>On October 27, 1999 at 14:28:15, James B. Shearer wrote:
>
>>On October 27, 1999 at 11:13:57, Amir Ban wrote:
>>
>>>On October 27, 1999 at 00:30:45, James B. Shearer wrote:
>>>
>>>>On October 26, 1999 at 12:54:02, Amir Ban wrote:
>>>>
>>>>>I got noplayed by crafty last week. I think it was playing more than 4 games in
>>>>>a row rather than disconnecting. At the time crafty's rating was 3146, so it
>>>>>seems it's rating jump is a result of noplaying ban. Crafty & clones typically
>>>>>gain around 200-300 points when they don't play other computers. I didn't
>>>>>realize crafty's noplay list is so long, but looking at it it effectively
>>>>>doesn't play top computers any more (ferret is hardly around these days).
>>>>
>>>> The bit about 200-300 points is hard to believe. In fact I don't
>>>>believe it. If this were true the crafty clones would be continuously pumping
>>>>rating points from the human pool to the computer pool. To maintain balance
>>>>some computers in the computer pool would have to be pumping just as many points
>>>>back into the human pool. Which computers do you think are doing this? I've
>>>>been playing crafty clones to inflate my rating but apparently this is all wrong
>>>>and I should be playing some other computers. Which are they?
>>>> James B. Shearer
>>>
>>>Mofongo, eggsalad, counterplay, wyrm and others. Some of them don't exclude
>>>computers in their formula, but they noplay me, and have always done so. I don't
>>>know who else they're noplaying.
>>>
>>>To find them, simply look at the top of the blitz list for crafty clones. To
>>>find those that play with everyone, look much lower in the list.
>>
>> This is not responsive to the point I was making. Suppose crafty
>>clones A, B and C play with everyone and have rating 2700. According to you
>>they would have rating 2900-3000 if they just played humans. This means that
>>when A, B and C do play humans they will be picking up lots of rating points (as
>>their rating tries to move from 2700 to 2900-3000). For their rating to stay at
>>2700 they must be losing the rating points they are picking up from humans to
>>some of the computers they are also playing. Suppose they are losing points to
>>computers X, Y and Z. Then X, Y and Z must also be losing the rating points
>>they are winning from A, B and C back to the human pool (else the rating of X, Y
>>and Z would rise to the point that X, Y and Z stopped winning rating points from
>>A, B and C).
>> So the question I am asking is what are names of programs X, Y and Z
>>whose ratings would shoot up if they noplayed humans? If A, B and C actually
>>exist then conservation of rating points means X, Y and Z must exist also. As a
>>human I want to play X,Y and Z and grab some of those rating points.
>>
>
>Your argument is too complicated for me to follow. It sounds like you are
>assuming that ICC ratings are in some sort of equilibrium and drawing
>conclusions based on that.
Of course ICC ratings are roughly in equilibrium. That is how rating
systems work. Your rating goes up or down until the rating points you are
winning are balanced by the rating points you are losing. There is quite a bit
of random noise in ICC ratings because the large K factor (K=32) but this also
means ratings are being driven to equilibrium quickly. Your original statement
about crafty clones gaining 200-300 points by noplaying computers only makes
sense interpreted as a difference between an old and new equilibrium level.
This is even more true of groups of players as some of the noise will
wash away. The flow of rating points between humans and computers must be
approximately in balance. Otherwise the relative ratings of the two pools would
quickly adjust so as to bring the flow into balance.
So if some computers (low rated crafty clones) are consistently winning
rating points from humans others (but which?) must be consistently losing rating
points to humans.
>I don't think theres any sort of equilibrium there. For one thing, it's not
>closed, and players are always entering and exiting the system. Besides, rating
>averages show huge drifts over time. I think blitz ratings are drifting
>perpetually higher, and in three years we will see above 4000. Another thing
>wrong with your argument is that the lower rated crafties will not be popular
>with humans since they have alternatives with same version & CPU but much higher
>rating.
There may be some factors causing slow long term drifts but they are
weak compared to the force adjusting the ratings to balance the inflow and
outflow of rating points from fixed groups (such as computers) of players.
As for crafty clones I expect you have it backwards. Crafty clones
artificially boost their rating by restricting which humans (and time controls)
they will play. The lower rated crafties are probably more popular because they
are less picky.
>Your conclusion, that there are no free rides at ICC, is certainly false. Anyone
>who's interested enough can boost his rating several hundred points by careful
>opponent selection. There's one or two crafties that have done exactly that.
I made no such conclusion. I just doubted noplaying computers is an
effective way for a crafty clone to raise its rating. It is no secret that
computers are better relative to humans at fast zero increment time controls.
So Mofongo can boost its rating just by exclusively playing 3 0 games against
humans. Noplaying humans that are good at 3 0 against it will boost the rating
even more.
>>>Last year there were several top computers playing regularly on ICC, and the
>>>standard 'high' rating was in the 2900's. Now serious comp-comp competition is
>>>almost disappeared from ICC and crafty, the only one still around, has gone up
>>>around 200-300 points.
>>
>> Maybe crafty got a lot better and has driven the competition from the
>>field of battle.
>> James B. Shearer
>
>Maybe, but I can also get a much higher rating than I could hope for last year,
>so maybe not.
So why do you think the other top computers left?
James B. Shearer
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.