Author: Christophe Theron
Date: 20:40:53 11/10/98
Go up one level in this thread
On November 10, 1998 at 13:10:04, Detlef Pordzik wrote:
>On November 10, 1998 at 12:41:09, Christophe Theron wrote:
>
>>On November 10, 1998 at 00:14:21, Les Walker wrote:
>>
>>>On November 09, 1998 at 23:56:24, Detlef Pordzik wrote:
>>>
>>>>On November 09, 1998 at 08:59:19, Micheal Cummings wrote:
>>>>
>>>>>Looking at the lastest SSDF rankings I was wondering where CM5500 was, the only
>>>>>ones they have are CM5000 ranked 26th then CM4000 at 51st.
>>>>>
>>>>>Can someone please tell me why even though CM6000 is very new, why there is no
>>>>>mention to adding it in the December ratings and CM5500 was not tested at all.
>>>>>and that program has been out for a while. I find the SSDF ratings very
>>>>>incomplete. and mis-guiding.
>>>>>
>>>>>If someone can answer why the most bought chess program versions in the world
>>>>>are not tested then I would like to know why please.
>>>>
>>>>While printing the new SSDF list I recognized something similar and asked Mr.
>>>>Karlsson in a email concerning this today.
>>>>
>>>>I doubt, I'll get an answer.
>>>>" No autoplayer " is absolutely no reason for my opinion, not to test a brandnew
>>>>product of this value.
>>>>As weather de Koning nor Distributor Mindscape have the slightest lobby down
>>>>there, nobody will have a look at the settings, that make sense. So, if at all,
>>>>CM will be sent into games with the standard settings - of 1 MB Hash tables -
>>>>against progs who refuse to start below 48 MB.
>>>>But, as said, they didn't care the least about this in the CM 5.000 Vs - they
>>>>didn't even use CM 5.500 - why should they use CM 6.000 at all ?
>>>>
>>>>ELVIS
>>>
>>>
>>>In my opinion, if the Chessmasters were included in the list, and everything
>>>were set fairly, i.e. hardware, software settings, etc., Chessmaster 6000
>>>would be number one, with Chessmaster 5555 being 2nd or 3rd on the list.
>>>It is also my opinion that the Chessmasters are not included for this very
>>>reason. I believe it is a "smart" business decision.
>>>
>>>Kind Regards,
>>>
>>>Les
>>
>>Fortunately there is the SSDF, so that people doing the kind of unverified
>>assertion you are doing here can be taken as they deserve: not seriously.
>>
>>Anybody can claim the program they like would be #1. Especially when a new
>>version is released. It is the best time to claim "if they would test it, it
>>would be #1". So that nobody can really check.
>>
>>In case you didn't notice, there are a lot of stronger programs out there.
>>
>>You can question the SSDF results, as it has been done, and give concrete
>>evidence if you can. We are not blind enough to 100% trust this list.
>>
>>But even taking it with a grain of salt (or 2 grains if you want), CM is nowhere
>>near the first place. You can claim that De Koning has improved his program by
>>100 elo points or more, based on several impressive games played against you.
>>And so what?
>>
>>You'd better give evidence, if you can, by posting games played against strong
>>programs.
>>
>>About the default hash table size: CM5000 comes with ttable_size=20 by default.
>>As I understand, this means "20 bits of hash table indexes", and it could mean
>>1M bytes, or 1M ENTRIES (what is the size of an entry? 8, 12 or 16 bytes?). I
>>don't know about CM6000.
>>
>>Even if it is only 1M byte, as The King has a very slow nps, I think CM does not
>>suffer a lot from a small hash table. Of course it would be a little stronger
>>with more HT, but don't overestimate this.
>>
>>And BTW, if it is really an issue, it is up to Mindscape to provide the
>>necessary information about how to change the hash table size. As far as I know,
>>the only way to change this size is by editing the CM.INI file. Or did I miss
>>something?
>>
>>Anyway, I think Mindscape doesn't care about the SSDF. They know they wouldn't
>>be close to #1, and even if they were it wouldn't increase their sales.
>>
>>You should better be thankful to the SSDF for the job they do. What have you
>>been doing personnaly to help them achieve better results? Send them an email to
>>explain how to increase the HT? Provide games yourself?
>>
>>Or do you think your critisism and assumptions about SSDF unfairness is the best
>>you can bring to the computer chess world?
>>
>>
>> Christophe
>
>tt= 20 means 1 MEGABYTE of hashtables, Mr. Theron.
>
>How come, you know or think to know, what Midscape thinks ? Name it your
>personal suggestion, that looks far better !
They are free to react here.
>Besides, whichever customer asks about changing of the settings, wilöl get a
>concrete answer, BTW - it's fairly known in common, anyway.
>
>What do you insist the comrade to do ? Send SSDF testers a description how to
>change the tt in a prog ?
I invite him to volunteer to play SSDF games instead of making unverified
assertions and critizism against peoples that really don't deserve this.
In one hand you have the SSDF guys, spending hundreds of hours playing games on
their own computers, without getting any money from this.
On the other end, one guy comes and say: "hey, this is bullshit. This program
should be number one".
I generally don't react to this kind of futile post. Sometimes, depending on my
mood, I feel the need to put things back on their place.
Because this is just a slap in the face of many people. In the face of the SSDF
guys, and in the face of the programmers whose program have a better ranking.
So from time to time somebody has to stand up and say: what you say is
completely unverified, and other readers have to know this.
>Who's doin' what up there ? I think the majority has a well settleld overall
>education concerning the useage of chess progs. So if they don't need soemthing
>-it's advice of this kind.
You are right. Programs should be tested with their DEFAULT settings, unless the
programmer itself asks to change something.
>Concerning the overall quality of the engine, which you obviously doubt, I
>presume it was way before you entered this scene -
When do you think I entered the scene?
> when de Konings engine led
>the SSDF list with an ELO of 2.402. His board computer - with a relatively poor
>hardware - comparing to todays needs (!) is still ranked No 31 - about one
>quality level with Jun. 4 on 90 MHZ - and this was an - old - engine.
I'm not sure to understand all of what you say.
There is no doubt that The King is a nice engine. I have a great respect for De
Koning's work. This engine has several strong points, a nice playing style,
agressive, active, always trying to find a way to attack your king, and good in
space evaluation.
Ok, but don't pretend it would be #1. Or at least give evidence.
Every week on this board we read such assertions.
>I, of course, can never foresee what the prog would achieve, if participating,
>at all - but I predict around rank 6-10; which I consider as brilliant enough.
This is a different story, and you might be right in the ranking you mention.
>What I'd else like to know :
>Who is " we " ....are not blind enough to.... ?
>You speak for several people, do you ?
At least for all the people who are able to do simple maths. The SSDF list is
published with the associated confidence numbers, or if you prefer the expected
margin of error on each elo ranking.
Some of the rankings overlap, so the exact places are sometimes "fuzzy".
I don't believe serious programmers would trust the list 100% anyway. We are
talking about statistical data, so caution must be exercised here.
But this list is the BEST data we have. Better than easy assertions from some
people.
Christophe
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.