Hi Fuming,
Remi's CG2006 paper is published/released half an year earlier. CG2006 was on
2006/5/29-31 at Turin, Italy. Mogo's paper
(http://hal.inria.fr/docs/00/12/15/16/PDF/RR-6062.pdf) was released on 2006/11.
That's why Mogo's paper cited Remi's paper (please see reference [9] in this
report).
As for MCTS, except the UCB formula, can you point out what is the difference
between Remi's MCTS and Mogo's UCT? In my understand, UCT is exactly a
selection strategy of MCTS. This is how Chaslot classified in his paper
"Progressive Strategies for Monte-Carlo Tree Search".
Aja
----- Original Message -----
From: Fuming Wang
To: [email protected]
Sent: Friday, December 31, 2010 2:02 PM
Subject: [Computer-go] Fwd: News on Tromp-Cook ?
---------- Forwarded message ----------
From: Fuming Wang <[email protected]>
Date: Fri, Dec 31, 2010 at 1:50 PM
Subject: Re: [Computer-go] News on Tromp-Cook ?
To: Aja <[email protected]>
Hi Aja,
Remi and S. Gelly's paper both come out in 2006,and I just checked that they
did not reference each other. I just read Remi's paper again, and realized that
CrazyStone's tree search approach is actually different from the popular UCT
method. Similar to you, I haven't been able to get good results from the
popular UCT method, so I might try CrazyStone's method for a change. In Remi's
paper, CrazyStone is only having around 30% winning rate against Gnu Go 3.6,
and now Erica is winning world competitions,this actually proves that high
quality MC simulation (realized for the first time in MoGo) is more important
than tree search algorithms.
Best regards,
Fuming
On Fri, Dec 31, 2010 at 12:41 PM, Aja <[email protected]> wrote:
Hi Fuming,
The idea of improving the quality of simulation is more earlier, than
Mogo’s paper, in the Appendix A of Remi Coulom’s CG2006 paper “Efficient
Selectivity and Backup Operators in Monte-Carlo Tree
Search”(http://remi.coulom.free.fr/CG2006/CG2006.pdf):
The choice of a more clever probability distribution can improve the
quality of the Monte-Carlo estimation....
I am not sure if Remi was the first one proposing this concept in Computer
Go field, but Mogo definitely was not.
I was in Amstertam attending Computer Olympiad 2007, in the team of Chinese
chess program “Deep Elephant”. I played with Crazy Stone, Mogo and was very
surprised to see they beat me. Afterwards, Mogo’s paper is so easy to
understand/implement for me that trigger me to work on Computer Go. Indeed,
Mogo has huge contributions, especially in the popularization of MCTS. I don’t
mean to weaken or deny it, but just want to point out Crazy Stone’s great
contributions. In Erica, I use CrazyStone-like simulations successfully.
Mogo-type simulation almost does not help Erica at all.
If we want to numerate the strongest programs, we cannot forget Fuego(2010
UEC Cup winner) and MyGoFriend(Computer Olympiad 2010, 9x9 winner). For
academic progress, we cannot forget Crazy Stone. For practical development
usage, we cannot forget GnuGo and GoGui released by Fuego team. There were
really too many contributors in the past.
Happy New Years to all.
Aja
From: Fuming Wang
Sent: Friday, December 31, 2010 10:16 AM
To: Aja ; [email protected]
Subject: Re: [Computer-go] News on Tromp-Cook ?
This is certainly a good time to sit back and look at what got us here. The
following key ideas have been mentioned so far: UCB, MCTS, RAVE, Pattern and Go
knowledge during MC simulation.These ideas are all essential to a strong MC
based Go program.If we want to pick the most important idea that got us here, I
would say it is the realization that adding Go Pattern and Go Knowledge to MC
simulation can significantly improve the quality of board evaluation. This is
amount to the important sampling concept in MC integration, which is very
import for Monte Carlo applications in many fields. MC simulation with
importance sampling give us for the first time a reasonablly accurate
evaluation function for Go. UCB, MCTS, RAVE are certainly very important,
however, it is still possible that new approaches that can achieve good results
with just importantly samples MC simulation. So, I think MoGo is the most
important break-through.
Happy New Year, everyone!
Fuming
On Fri, Dec 31, 2010 at 9:20 AM, Aja <[email protected]> wrote:
Hi Jeff,
When, do you think, did Mogo "started dominating all the KGS computer
events and CGOS, and also was the first to extend that dominance from 9x9 to
19x19."?
In Computer Olympiad 2007, Steenvreter was gold medal on 9x9. At the
final match of 19x19, it's easily to see that Mogo and Crazy Stone were close
(finally Mogo 1st and CS 2ed). But, at the end of 2007, Crazy Stone defeated
Mogo and won the UEC Cup (19x19). Afterwards, Many Faces won 9x9 and 19x19 on
2008. Zen and Erica won 2009 and 2010, both continuing Crazy Stone thread.
Mogo's biggest contributions, so far, in my view, are
1.Applied UCT to computer Go, and such application came from the idea
"MCTS" that proposed in 2006 by Remi Coulom. Crazy Stone was using MCTS to win
9x9 in 2006 Computer Olympiad.
2.See 3x3 patterns around the previous move.
3.RAVE (strictly speaking, it is invented by David Silver).
UCT and RAVE are for both for the tree search. I think Crazy tone's
contribution for the playout is of same/or more important, because the quality
of simulations decide the playing strength much. >From this view, we should
give Crazy Stone more and more credit.
I don't mean to raise any debate. Mogo does has important contributions,
but it's not so hard to assign credit to Crazy Stone. By the way, we should not
forget Fuego and MyGoFriend. Anyway, I think SenSei's description is
out-of-date.
Aja
-----原始郵件----- From: Jeff Nowakowski
Sent: Friday, December 31, 2010 5:43 AM
To: [email protected]
Subject: Re: [Computer-go] News on Tromp-Cook ?
On 12/30/2010 01:58 PM, David Fotland wrote:
You should also give more credit to CrazyStone as an early strong
program
that contributed many ideas, comparable to Mogo. Remi is Aja's
advisor, so
Erica continues the CrazyStone thread.
I did mention CrazyStone, and the Sensei's page lists it first as the
program that "started the new wave of MCTS programs by winning the 9x9
gold medal at the ICGA Computer Olympiad, in 2006." Like I said in my
first message, though, it's hard to assign credit, and I don't mean to
slight other programs.
However, MoGo was the program that really got people to sit up and take
notice, because it started dominating all the KGS computer events and
CGOS, and also was the first to extend that dominance from 9x9 to 19x19.
I believe the biggest breakthroughs were made with MoGo (building, of
course, on earlier ideas). This is easily verified by going back to the
archives and seeing how many people patterned their program after MoGo.
_______________________________________________
Computer-go mailing list
[email protected]
http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
_______________________________________________
Computer-go mailing list
[email protected]
http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
------------------------------------------------------------------------------
_______________________________________________
Computer-go mailing list
[email protected]
http://dvandva.org/cgi-bin/mailman/listinfo/computer-go_______________________________________________
Computer-go mailing list
[email protected]
http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
