Re: Chess Clocks: pet peeves and preferences
raylopez99
> Things I dislike: All clocks:
Some other thread ideas: please tell us what you don't like about
computer chess, playing online, and finally, chess itself.
Chess versus Go--which is more played out? I say Go. Those players
are arrogant and really think a PC cannot learn Go--they'll find their
comeuppance too, just like chess players 20 years ago, since Go
seems to be about that far behind in its development vis-a-vis chess.
RL
Mike Murray
<raylo...@gmail.com> wrote:
>On Feb 17, 9:19�pm, Mike Murray <mikemur...@despammed.com> wrote:
>> Things I dislike: �All clocks:
>
>Some other thread ideas: please tell us what you don't like about
>computer chess, playing online, and finally, chess itself.
Where do you play online? (besides get-club, I mean). What's yer
handle?
Michael Alford
TROLL ALERT!
Michael Alford
and by-the-by, I quit playing chess 40 years ago when I started playing
Go, but even after all that time I know it's ruy lopez
fuck off, troll
Joel Olson
news:4b7c7265$0$86454$39ce...@news.twtelecom.net...
>
Soon, we'll hear from Roy Mopeds.
Robert Jasiek
>are arrogant and really think a PC cannot learn Go--they'll find their
>comeuppance too, just like chess players 20 years ago, since Go
>seems to be about that far behind in its development vis-a-vis chess.
(Having a possibly flawed prediction of the future is not arrogance
but is restricted foresight.)
Go programs were delayed in their development for several reasons:
- During the mid-20th century, Go playing strength of players outside
East Asia was still comparatively moderate and there were considerable
fewer players than today. So at that time only few go programmers
existed.
- Go has a strategic development different from Chess. Go is more
strategy-heavy while Chess is more tactics-heavy. AI suitable for
especially strong Go playing programs had not been available until
recently.
- For a long time, Go programs followed the AI expert system path. As
it has turned out, this approach requires more research and manpower
than could be invested so far for really strong programs. A suitable
path for Go using the computers' advantage (sheer calculation power)
was found / recognized only about a decade ago. Successful
implementations were produced still a few years later (I think around
2003). With them, current Go programs have a playing strength roughly
comparable to Chess programs of the 50ies.
- Combining different approaches (MonteCarlo-like playouts with more
expert knowledge than at the moment) will make Go programs stronger
(from currently about amateur 3 dan on small boards to possibly high
dan level on the 19x19 board). Maybe within the next 50 years.
- Usage of cluster computer networks for Go has been the exception so
far. There is no regular Deep Blue hardware of any kind in the Go
programs scene.
Go programs relying on brute force (like MC) do not explain Go to
human players though. It requires different kinds of research to
increase the level of human play.
micky
.
> and by-the-by, I quit playing chess 40 years ago when I started playing
> Go, but even after all that time I know it's ruy lopez
No ! it's treenie bumsuck !
raylopez99
> On Wed, 17 Feb 2010 12:06:56 -0800 (PST), raylopez99
>
> >On Feb 17, 9:19 pm, Mike Murray <mikemur...@despammed.com> wrote:
> >> Things I dislike: All clocks:
>
> >Some other thread ideas: please tell us what you don't like about
> >computer chess, playing online, and finally, chess itself.
>
> Where do you play online? (besides get-club, I mean). What's yer
> handle?
That's confidential. I play at the Class A level maybe Expert if I
get lucky, but online I'm a different animal than on Usenet.
And you?
RL
raylopez99
> and by-the-by, I quit playing chess 40 years ago when I started playing
> Go, but even after all that time I know it's ruy lopez
>
> fuck off, troll
And what have you learned in 40 years of playing GO? Nothing.
Because Go is like Pachinko--the chances of winning are largely
determined by chance between two players of roughly the same skill.
No skill in other words, instead it's like a coin toss.
By contrast, chess involves real skill and the player that loses lost
for a reason that can be pointed to--not like the mush that's Go.
GO TO H***, Michael Alford.
RL
raylopez99
> Go programs relying on brute force (like MC) do not explain Go to
> human players though. It requires different kinds of research to
> increase the level of human play.
Thanks Robert for that scholarly assessment of GO software at the
moment. When you say "expert system" for GO is it heuristic or rules
based or both? In short does the program "learn" from pattern
recognition (heuristic) or from IF/THEN rules?
RL
Robert Jasiek
<raylo...@gmail.com> wrote:
>When you say "expert system" for GO is it heuristic or rules
>based or both?
It can be either or both. Different programs have used different
approaches. Heuristics combined with search techniques, hashing etc.
were the most frequent before the MC age. I think a very few programs
used rules to express heuristics but apparently not very consistently.
>In short does the program "learn" from pattern
>recognition (heuristic) or from IF/THEN rules?
Real learning programs were scarce. Mostly learning was specialized to
a single task like, e.g., patterns. A few used learning from game
databases, with rather little success.
Even fewer experimented with neural nets (for specialized tasks) to no
avail. - The most eccentric learning was a population of neural nets
trying to kill each other or survive in a simulated evolution. Those
"critters" first understood that passing was not that good, later
found some value in making eyes but never went beyond that. The
programmer's wife commented: "It is like the real evolution: You will
have to wait for billions of years!" ;)
So most "learning" in expert systems was (and still is) endlessly
changing the hard source codes... But MC and its successors is
changing everything now.
Mike Murray
I play under my real name," MikeMurray", on playchess. Unless I'm
playing bullet. Then it's "Jerry Attrich".
Is the reason your handle is confidential itself confidential ?
Michael Alford
>
> And what have you learned in 40 years of playing GO? Nothing.
> Because Go is like Pachinko--the chances of winning are largely
> determined by chance between two players of roughly the same skill.
> No skill in other words, instead it's like a coin toss.
you are an ignorant fool, and bedamned troll, you've pulled this crap before
raylopez99
> On Thu, 18 Feb 2010 11:15:35 -0800 (PST), raylopez99
>
> >
> So most "learning" in expert systems was (and still is) endlessly
> changing the hard source codes... But MC and its successors is
> changing everything now.
MC standing for??? MicroController? That would be too obvious.
RL
raylopez99
>
> I play under my real name," MikeMurray", on playchess. Unless I'm
> playing bullet. Then it's "Jerry Attrich".
>
> Is the reason your handle is confidential itself confidential ?
Yes, very.
playchess...rings a bell...that's ChessBase's site. I was paying but
now I'm just a Guest. FICS is another free favorite of mine, play a
bit stronger than playchess it seems. Don't know about ICC, but I'm
not so diehard to pay to play.
RL
raylopez99
You can't prove that. And you cannot answer the question--why GO is
just a game of chance. Actually it's a trick question--because at the
highest level between even players, even chess is a game of chance.
Anyway I like chess because you can wrap up a game in 40 moves--not
40000 like Go that lasts for weeks on end.
Baseball vs. cricket--which is better? like Chess vs. Go. Chess is
better.
RL
Taylor Kingston
> On Feb 19, 1:06 am, Michael Alford <m...@aracnet.com> wrote:
>
> > you are an ignorant fool, and bedamned troll, you've pulled this crap before
>
> You can't prove that.
Based on my experience, I'd say he's right, Ray.
Robert Jasiek
<raylo...@gmail.com> wrote:
>MC standing for???
As said before, MonteCarlo.
http://en.wikipedia.org/wiki/Monte_Carlo_method
http://senseis.xmp.net/?UCT
>Anyway I like chess because you can wrap up a game in 40 moves--not
>40000 like Go that lasts for weeks on end.
Ca. 230 is the average. The record in a real game is about 450. The
theoretical problem record is about 19.6 million.
Go can be played also in 5 or 10 minutes sudden death lightning, if
desired.
raylopez99
> As said before, MonteCarlo.http://en.wikipedia.org/wiki/Monte_Carlo_methodhttp://senseis.xmp.net/?UCT
I fail to see how Monte Carlo simulations will determine where to
place stones on a grid, unless you are basing it on some sort of
pattern recognition where it's been found that this shape X will morph
into a winning shape Y (say an "island" or "eye" or whatever) with a
certain probability. Then, you can compute each legal move in a
certain position and, like cloud identification with storm activity,
figure out which paths are most likely to lead to a winning "pattern"
or shape Y.
>
> >Anyway I like chess because you can wrap up a game in 40 moves--not
> >40000 like Go that lasts for weeks on end.
>
> Ca. 230 is the average. The record in a real game is about 450. The
> theoretical problem record is about 19.6 million.
>
> Go can be played also in 5 or 10 minutes sudden death lightning, if
> desired.
I see.
RL
raylopez99
wrote:
> > On Feb 19, 1:06 am, Michael Alford <m...@aracnet.com> wrote:
>
> > > you are an ignorant fool, and bedamned troll, you've pulled this crap before
>
Your experience is limited Taylor. Recall our debate and your
demonstrated ignorance of the principles of Neo-Lamarkian inheritance
in the cichlids of Lake Tangayika.
That spoke volumes.
RL
Robert Jasiek
<raylo...@gmail.com> wrote:
>I fail to see how Monte Carlo simulations will determine where to
>certain position and, like cloud identification with storm activity,
>figure out which paths are most likely to lead to a winning "pattern"
>or shape Y.
It is done essentially independently of patterns. Simply consider move
X, run 100,000 sequences until their game ending passes, where the
position's score is given, then you know the percentage of sequences
yielding a win. Do this for each legal move X in the current position
and choose the move with the greatest probability.
MC programs really play absolutely stupid, legal, randomly selected
sequences. But it works. Well almost. Some tweaking is being done to
modify the basic idea. Like storing only(!) 8 patterns on a 3x3
subboard that are preferred to pure random choice.
Advanced MC uses iterative depening and other methods, but this is
engineering rather than clean theory.
Michael Alford
Yup, a cursory glance at Google shows posts by raylopez99 to threads
such as "Chess is Superior to Go" and "Computer Science Experts: Chess
'harder' than Go".
Hans-Georg Michna
>On Fri, 19 Feb 2010 03:35:33 -0800 (PST), raylopez99
><raylo...@gmail.com> wrote:
>>I fail to see how Monte Carlo simulations will determine where to
>>place stones on a grid [...]
>[...] MC programs really play absolutely stupid, legal, randomly selected
>sequences. But it works. Well almost. Some tweaking is being done to
>modify the basic idea. Like storing only(!) 8 patterns on a 3x3
>subboard that are preferred to pure random choice. [...]
MC playout as a means to evaluate the position has brought about
a fairly sudden jump in computer go capability, but I believe it
will ultimately be a dead-end street. This alone, even with a
lot more raw power, may never reach the higher dan levels.
I guess we need another algorithmic idea or perhaps a truly huge
jump in computer performance.
Of course I could be mistaken. Just a gut feeling.
Hans-Georg
raylopez99
> It is done essentially independently of patterns. Simply consider move
> X, run 100,000 sequences until their game ending passes, where the
> position's score is given, then you know the percentage of sequences
> yielding a win. Do this for each legal move X in the current position
> and choose the move with the greatest probability.
I see. I did not realize that in Go you can have a "score" like in
chess, based on the number of stones I suppose. So you can use the
famous "alpha-beta" algorithm of chess, in a tree? Such as, if I move
there, he moves there, and for that position he beats me so I can
prune the tree? I don't think it can be that easy (otherwise the
problem would have been solved), so it must be that the games are
played to the end of the line (when no legal moves exist). In which
case, your N-ary tree becomes very huge, very very fast, since no
pruning of bad moves allowed until you reach the 'end' or 'leaves' of
the tree (i.e, in chess, you can capture the queen, but if your
opponents next move can be a mate, then capturing the queen is a bad
move and hence that move can be 'cut' or 'pruned' from the list of
plausible moves, making finding the right move in the chess tree much
easier--I take it in Go the tree cannot be so easily pruned).
RL
raylopez99
> Yup, a cursory glance at Google shows posts by raylopez99 to threads
> such as "Chess is Superior to Go" and "Computer Science Experts: Chess
> 'harder' than Go".
Proves nothing. "Cursory" glance--noted. It could be somebody else
used this computer. It could be chance. Proves nothing.
RL
Robert Jasiek
<raylo...@gmail.com> wrote:
>So you can use the famous "alpha-beta" algorithm of chess, in a tree?
It or other search algorithms are used especially for expert system go
programs, too.
>I take it in Go the tree cannot be so easily pruned).
Right.
Robert Jasiek
<hans-georgN...@michna.com> wrote:
>MC playout as a means to evaluate the position has brought about
>a fairly sudden jump in computer go capability, but I believe it
>will ultimately be a dead-end street. This alone, even with a
>lot more raw power, may never reach the higher dan levels.
>
>I guess we need another algorithmic idea or perhaps a truly huge
>jump in computer performance.
>
>Of course I could be mistaken. Just a gut feeling.
It might also be that a relatively small number of expert system
addons will suffice together with slightly increased calculation
power:
- ladders
- avoiding tooth paste techniques
- no overevaluation of premature removals
- no premature wasting of ko threats
etc.
Hans-Georg Michna
The problem here is that such algorithms slow down the
calculation quite a bit, so they often waste more than they
yield. They can easily be slower than blind MC playout and may
not usually be much better.
Besides, aren't these obvious enough that programmers would have
tried them? My best guess is that some have and found them not
beneficial.
But we cannot really discuss this in any more detail, as it
remains mere guesswork. I have never written a go-playing
program.
Hans-Georg
Robert Jasiek
<hans-georgN...@michna.com> wrote:
>The problem here is that such algorithms slow down the
>calculation quite a bit, so they often waste more than they
>yield. They can easily be slower than blind MC playout and may
>not usually be much better.
>
>Besides, aren't these obvious enough that programmers would have
>tried them? My best guess is that some have and found them not
>beneficial.
Yes, it slows down MC and some if it has been tried (ko threats
probably not yet). So currently MC programs become weaker when trying
to consider ladders etc. by means of exceptions. It will be the task
of future research how to achieve it that added knowledge actually
helps (instead of hurting) otherwise rather pure brute-force
approaches. Understanding of such interdependent and somehow
competitive algorithmic apporaches is in its infancy.
rt
>
> On Feb 19, 1:45 pm, Robert Jasiek <jas...@snafu.de> wrote:
> > It is done essentially independently of patterns. Simply consider move
> > X, run 100,000 sequences until their game ending passes, where the
> > position's score is given, then you know the percentage of sequences
> > yielding a win. Do this for each legal move X in the current position
> > and choose the move with the greatest probability.
>
> I see. I did not realize that in Go you can have a "score" like in
> chess, based on the number of stones I suppose. So you can use the
The merest bit of research on the game would have revealed that...
rt
Do you think it was Elvis? Or perhaps some weird entropic happening?