Author: Christophe Theron
Date: 15:20:19 04/09/02
Go up one level in this thread
On April 09, 2002 at 17:26:24, Will Singleton wrote:
>On April 09, 2002 at 15:05:59, Christophe Theron wrote:
>
>>On April 09, 2002 at 09:31:49, Will Singleton wrote:
>>
>>>On April 08, 2002 at 23:14:40, Christophe Theron wrote:
>>>
>>>>On April 08, 2002 at 21:25:46, Will Singleton wrote:
>>>>
>>>>>On April 08, 2002 at 20:54:35, Jorge Pichard wrote:
>>>>>
>>>>>>On April 08, 2002 at 20:23:57, Will Singleton wrote:
>>>>>>
>>>>>>>
>>>>>>>Wow, ugly game. I'm having a hard time understanding this, it seems illogical.
>>>>>>>Are you playing many games, then selecting the one Sjeng loses? If so, how
>>>>>>>many?
>>>>>>
>>>>>>I have played many games against other winboard programs such as Gaviota,
>>>>>>Bestia, faile, Movei, and even against Novag Turqouise; but against Sjeng so far
>>>>>>only three and the first one I did NOT included since I was using 12 MB default
>>>>>>for my Celeron 433 Mhz and my Celeron only had 32 MB of memory up until
>>>>>>yesterday when I decided to upgrade it to 128 MB.
>>>>>>
>>>>>>
>>>>>>I really don't have much time to test CTP 14.9 since like most of us I do have a
>>>>>>job. And to answer your silly question, I don't select games, as a matter of
>>>>>>fact I was as dubious as you are now, and before I had Chess Tiger for Palm I
>>>>>>also was a disbeliever to even consider CTP to be close to being of an expert
>>>>>>strength. The only reason why I continue to test CTP 14.9 is probably because I
>>>>>>resemble you in a way, I'm still NOT too conviced that such a little Gizmo can
>>>>>>accomplish so much against programs that are considered close to 2400 in
>>>>>>strength, but the more I test this Little Beast, the more I have to accept the
>>>>>>fact that it is amazinly strong.
>>>>>>
>>>>>>Pichard.
>>>>>
>>>>>Look, I appreciate that you test and post games. My question was prompted by
>>>>>comparing your posted game with sjeng 12L, which seemed to avoid several poor
>>>>>moves (though I must admit, it did play BxN with the fiachettoed bishop, then
>>>>>went on to destroy its own kside).
>>>>>
>>>>>I don't think my question was silly at all. Had you posted your methodology, I
>>>>>wouldn't have had to question you. It's a reasonable thing to ask.
>>>>>
>>>>>I don't believe CT for Palm is a strong program, simply due to its limited
>>>>>search depth.
>>>>
>>>>
>>>>You talk about methodology in your second paragraph. Good point.
>>>>
>>>>What methodology have you used to evaluate the search depth of Chess Tiger for
>>>>Palm and to decide that it is "limited"?
>>>>
>>>>Do you have a Palm? Have you tested Chess Tiger for Palm? I can't find you in my
>>>>customer's list.
>>>>
>>>>
>>>>
>>>> Christophe
>>>>
>>>>
>>>
>>>I have been using and testing CT for Palm since you brought out 14.1 text-mode
>>>(I believe that first version was fully functional, though I don't have it
>>>anymore). I did a lot of testing with it, ran some epd test suites, played
>>>games vs Palm Genius, played against my program, and played it some myself.
>>
>>
>>There has been a lot of changes in the search algorithm since 14.1. The current
>>version is 14.9, one additional year of research.
>>
>>
>>
>>> CT
>>>for Palm, imho, is certainly a very fine program.
>>
>>
>>Thanks.
>>
>>
>>
>>>Common sense, as well as the testing I've done, says that CTpalm will search
>>>less deeply than most amateur pc programs. That's what I meant by "limited"
>>>search depth. The fact we're discussing it at all is a tremendous compliment to
>>>CTpalm, and to your dedication and programming skill. I will shortly add myself
>>>to your database, and recommend that others do so. :)
>>
>>
>>Great.
>>
>>
>>
>>>Having said that, a few more comments. There has been some discussion in the
>>>past regarding the playing strength of the Novag Sapphire II vs chess programs
>>>on the Palm. I read a comment by the fellow who automated the Sapphire on fics,
>>>who said that his account is routinely crushed by any reasonable pc program
>>>running on a P166 or greater (Sapphire runs at 32mhz, I think). In testing
>>>Sapphire vs CTpalm, he also reported CT was getting outsearched by a ply or so
>>>in a few test games, and did not win any.
>>
>>
>>That's not what has been reported to me, but I guess we do not have enough data
>>to be sure.
>>
>>
>>
>>> Put those two observations together.
>>>Now, admittedly that’s blitz, and standard time-control might be different.
>>>However, you said yourself that CT should be about equal in strength to the
>>>Sapphire, citing the Sapphire’s SSDF rating. You also said that the Sapphire
>>>should be “wiped away” by any program running on a Strongarm at 100mhz. Put
>>>those two observations together. Given these and other reported results here,
>>>you can readily understand my scepticism at the Sjeng games. However, strange
>>>things do happen.
>>
>>
>>I think Jorge's computer has some problems and I'm wondering how much this is
>>affecting the results.
>>
>>However he is not the only one to test Tiger for Palm, and "strange" results are
>>coming from several different sources now.
>>
>>
>>
>>
>>>As for testing methodology, you're right that I haven't performed rigorous
>>>tests. But that's tough to do, since all games have to be run manually, and at
>>>standard time control (since, as you have said, CTpalm doesn't do well at
>>>blitz). I think a serial interface between Winboard and Palm would be necessary
>>>to perform reasonable testing. I'd be happy to write the pc side of the
>>>interface. I guess that would require you to add some interface code on the
>>>Palm side, or perhaps you already have an interface.
>>
>>
>>No I do not have an interface, and I'm not planning to add one soon.
>>
>>The problem is the lack of standard. I could write a specific interface, but it
>>would work only with a given class of opponents (for example PC programs running
>>under Winboard). That's too much work for a limited result.
>>
>>
>>>I enjoy CT for Palm, as I do other programs. And I like to test them,
>>>especially against mine. CT can do very well, no doubt about it. But, let’s
>>>not get carried away and imply that it approaches amateur pc programs in
>>>strength.
>>
>>
>>It does. It does not approach the best ones, but it approaches some of the
>>average ones, which is useful to estimate its strength.
>>
>>I think it also shows that hardware is not the only dominant point in computer
>>chess. It's hard to come up with good search algorithms, but they really make a
>>hell of difference with "standard" or "public" ones.
>>
>>Maybe some people have forgotten about this (which is the reason why we see the
>>words "surprise" or "miracle"), so the experiment was worth it.
>>
>>BTW it has been a little surprise for me as well. But it's refreshing. It
>>reinforces my faith in "software".
>>
>>
>>
>> Christophe
>
>
>You say that CT approaches the average amateur programs in strength. I should
>probably defer to your assessment, since you have the most experience with it.
>But this is my problem: there's no hard data to either support or refute your
>claim. We can only get data with extended and repeatable testing, which people
>really can't do without automation. So we're stuck.
>
>Will
I think that on a fast Palm Chess Tiger is 2200 SSDF elo. That's the strength of
many amateur progams, approximately.
So yes, I believe that CT for Palm approaches the strength of the average
amateur program.
Yes, approximately.
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.