Re: [computer-go] Modern brute force search in go

[Vincent Diepeveen <diep@xxxxxxxxx>]

At 17:33 7-11-2004 -0500, Don Dailey wrote:
>
>
>>   d:  1 s:     3 n:        365 t:     0.4s pv: s18
>>   d:  2 s:     1 n:       1812 t:     0.5s pv: s18 s19
>>   d:  3 s:     4 n:       8743 t:     0.7s pv: r18 s18 o18
>>   d:  4 s:     0 n:     249878 t:     6.6s pv: s18 s3 s7 b18
>>   d:  5 s:     4 n:     638195 t:    18.0s pv: s18 r18 s3 q3 s15
>>   d:  6 s:     0 n:   13871970 t:   377.2s pv: s18 r18 s15 t15 b18 o18
>>   d:  7 s:     4 n:   57710143 t:  2039.0s pv: s18 r18 s15 t15 s3 s7 b18
>
>Does your program actually benefit from null move or does it just search
>more selective?

As GCP just had to catch a plane, i'll answer that for him.

GCP's thing is probably the only go program where my own poor go program
can win from, if i still can find its source code somewhere.

I'm sure GCP's thing doesn't play better even when getting 100 ply.

It doesn't know about liberties of a group. Let alone that it removes
captured groups off the board.

I'm keeping track of all that incremental of course.

The problem of computerchess is a number of things.


No one plays with those first few versions of diep. they were 'shareware'.
Chessmaster, and that has a very strong engine called The King, sold over 6
million copies already years ago. Fritz has no public statistics, but i
would be guessing now that they get sold in so many shops now they must be
having sold like 2 million copies of fritz. 

Diep has been at 10+ million cdroms and i didn't get a penny for that. It
had a nice graphical interface and it was kind of freeware for a while. It
begged for money, but no one out of that 10 million ever registered it in
the early days. That only happened when it was privateware/commercial. 

So the total market is just huge. 

There is many yelling researchers and professors in computerchess who guess
they know everything, but in fact they are just busy letting themselves
look good and in case of american professors they protect the american
programs a lot too by saying nothing ever will get better than that (like
deep blue), whereas the computerchess gets dominated by 1 israeli program,
1 german program, and a bunch of dutch programs. Also very strong now is 1
austrian program, playing under the flag of United Arab Emirates, as the
sheikh owns it now. It is called Hydra.

In short if i draw a circle from here with r=100 kilometer i have already 3
out of the first 4 finishers in world champs 2004 (shredder,diep,fritz).

Amazingly crafty has become 5th somehow in world champs 2004. It doesn't
have anything new though. It always runs behind, but it is well tested and
tuned and plays bugfree and this year had a very clever operator who also
played the openingsmoves (they are very important in chess). 

The first line crafty prints out is: "Hello from Crafty\n"

There is incredible number of persons X who have their 'own' program which
at the first line says: "Hello from X\n".

People like to win.

With chessprograms you can win games from titled players.

You will get that all in go in a year or 15 too.

As i'm a titled player, there have been times at chess servers that every
unregistered player who wanted to challenge me, which happened each few
minutes, was a hidden computer.

Many titled players have really grown sick of that.

People just like to win.

In chess they can.

In fact many registered players are computers too. 2 days ago i blitzed a
little on a free chess server (FICS) and i played again some registered
player who was clearly cheating with a computer.

This despite all the 'cheater catching software'.

In go this is more unlikely to happen. Because even untitled players there
already nowadays never accept a game from a very weak player, let alone
they chat to you when you ask what you did do wrong during the game.

Every long chess game i play in my life, afterwards i analyze with my
opponent. 

Isn't that the habit in go?

The reason why all computerchessprogrammers like Go, is because it is not
mature yet in search so all kind of sophisticated forward pruning
techniques can get invented there still and the openings book is not so
important in go.

Sometimes in chess you just lose a game because your opponent prepared the
opening better. It is this far that a computerchess program cannot do the
openingspreparation himself. It is a fulltime job. Additionally there is
more good programs than there are good openingsbookers.

The progress in openingsbooks last few years is scary. In 1997 i had an
openingsbook of 5000 hand given in moves. Deep Blue had 4000 hand given in
moves. 

In 2004 the top programs all have books of 2+ million moves from which
majority is hand given in.

Go has that problem to lesser extend.

However when the go programs get a lot stronger in a year or 15, i would be
amazed if not the similar things are going to happen there too.

In Go you don't need to prepare 1 certain opening, but you can figure out
which type of openings build up is the achillesheel of a competitor X.

The word 'openingsbook' can be exchanged for 'openingstuning' then.

Oh dear now i probably gave at least 1 commercial programmer a plan :)

Make that 15 days :)

>- Don
>_______________________________________________
>computer-go mailing list
>computer-go@xxxxxxxxxxxxxxxxx
>http://www.computer-go.org/mailman/listinfo/computer-go/
>
>
_______________________________________________
computer-go mailing list
computer-go@xxxxxxxxxxxxxxxxx
http://www.computer-go.org/mailman/listinfo/computer-go/