It is currently Thu Mar 28, 2024 6:00 am

All times are UTC - 8 hours [ DST ]




Post new topic Reply to topic  [ 76 posts ]  Go to page Previous  1, 2, 3, 4
Author Message
Offline
 Post subject: Re: DeepZenGo thread
Post #61 Posted: Sat Nov 18, 2017 12:38 pm 
Lives in gote

Posts: 474
Liked others: 120
Was liked: 157
Rank: igs 4d+
Deepzengo is losing a lot these days against a player named bus89. I only did a quick check so I may be mistaken but it seems bus89 won the last 9(!)games he played against Deepzengo. Does someone know who bus89 is?

Here are the last three games, the two games with bus89 as black are identical up to move 47 (and this opening is also found in other games where bus89 played black, and won, against Deepzen).







Attachments:
bus89(P)(9D)_DEEPZEN(B)(9D)_201711181554.sgf [1.28 KiB]
Downloaded 1004 times
DEEPZEN(B)(9D)_bus89(P)(9D)_201711181721.sgf [1.29 KiB]
Downloaded 1047 times
DEEPZEN(B)(9D)_bus89(P)(9D)_201711181842.sgf [1.32 KiB]
Downloaded 1090 times

This post by Shenoute was liked by: Bonobo
Top
 Profile  
 
Offline
 Post subject: Re: DeepZenGo thread
Post #62 Posted: Sat Nov 18, 2017 5:38 pm 
Judan

Posts: 6725
Location: Cambridge, UK
Liked others: 436
Was liked: 3719
Rank: UK 4 dan
KGS: Uberdude 4d
OGS: Uberdude 7d
Interesting that Zen is still playing the cut that doesn't work (according to human analysis) in the micro Chinese opening. I remember when Zen first got 8 and then 9 dan on KGS about a year ago it did this and several people exploited it to get an early lead. However Zen sometimes managed to win anyway by dying but squeezing cut outside groups.

Top
 Profile  
 
Offline
 Post subject: Re: DeepZenGo thread
Post #63 Posted: Sun Nov 19, 2017 6:20 am 
Lives in gote

Posts: 474
Liked others: 120
Was liked: 157
Rank: igs 4d+
Bus89 also seems to have found a "standard" opening when playing white against Deepzen. The first 44 moves of their last 4 games (with bus89 playing white) were as follows:



The opening itself is not wonder weapon though since before winning these 4 games with white, bus89 had lost one game playing it.
I also saw that other pros are apparently playing these openings against Deepzen. KeA and Bahamut won today as black, playing the opening shown in my previous post.


Attachments:
bus89-Deepzengo (Tygem).sgf [414 Bytes]
Downloaded 989 times
Top
 Profile  
 
Offline
 Post subject: Re: DeepZenGo thread
Post #64 Posted: Sun Nov 19, 2017 6:47 am 
Lives in gote

Posts: 474
Liked others: 120
Was liked: 157
Rank: igs 4d+
A third "standard" opening against Deepzen seems to be in the making (up to move 34):





I don't know if pros are really making progress against AI but all of Zen's seven losses on the 19th featured one these three openings.


Attachments:
XIUZHI(P)-DEEPZEN(B)_20171119 (Tygem).sgf [1.26 KiB]
Downloaded 1000 times
wjx(P)-DEEPZEN(B)_20171119 (Tygem).sgf [2.39 KiB]
Downloaded 986 times
Top
 Profile  
 
Offline
 Post subject: Re: DeepZenGo thread
Post #65 Posted: Sun Nov 19, 2017 8:06 am 
Judan

Posts: 6725
Location: Cambridge, UK
Liked others: 436
Was liked: 3719
Rank: UK 4 dan
KGS: Uberdude 4d
OGS: Uberdude 7d
Shenoute wrote:
Bus89 also seems to have found a "standard" opening when playing white against Deepzen. The first 44 moves of their last 4 games (with bus89 playing white) were as follows:


When this (new, formely considered white overplay) joseki appeared in one of the Master games the conclusion was the p13 push was a mistake and better to directly defend the corner with empty triangle as the cutting stones can't be captured and you don't know whether to turn or push yet. That situation was a little less open though.


This post by Uberdude was liked by: Shenoute
Top
 Profile  
 
Offline
 Post subject: Re: DeepZenGo thread
Post #66 Posted: Sun Nov 19, 2017 2:15 pm 
Lives in gote

Posts: 474
Liked others: 120
Was liked: 157
Rank: igs 4d+
Two more games lost by Deepzen today. One played on Tygem against bus89, in which bus89 had black and scored a win using the mini-chinese based opening mentioned above. The other comes from WBaduk and is a win by Shuto Shun, a Japanese 7p (the shape/sequence in the upper left was commented upon by Uberdude some time ago if I remember well).





Attachments:
DEEPZEN(B)(9D)_bus89(P)(9D)_201711191648.sgf [1016 Bytes]
Downloaded 999 times
201711191749deepzengoshutoshun54.sgf [1.28 KiB]
Downloaded 948 times
Top
 Profile  
 
Offline
 Post subject: Re: DeepZenGo thread
Post #67 Posted: Sun Nov 19, 2017 2:52 pm 
Lives in sente

Posts: 727
Liked others: 44
Was liked: 218
GD Posts: 10
Human strikes back! But v.15.6 or 15.5-0.6 release today so I'm observing.

Top
 Profile  
 
Offline
 Post subject: Re: DeepZenGo thread
Post #68 Posted: Sun Nov 19, 2017 3:33 pm 
Lives in gote

Posts: 474
Liked others: 120
Was liked: 157
Rank: igs 4d+
pookpooi wrote:
Human strikes back! But v.15.6 or 15.5-0.6 release today so I'm observing.

The last few days might have been the last high point for humans :-) It will be interesting to see if the new Zen can deal with the openings above.

Edit. I just checked on Tygem, Deepzen has played two games after a two hour break. Two losses with black against bus89.

Top
 Profile  
 
Offline
 Post subject: Re: DeepZenGo thread
Post #69 Posted: Tue Nov 21, 2017 8:33 am 
Lives with ko

Posts: 136
Liked others: 47
Was liked: 21
Rank: KGS 6 dan
Uberdude wrote:
Interesting that Zen is still playing the cut that doesn't work (according to human analysis) in the micro Chinese opening. I remember when Zen first got 8 and then 9 dan on KGS about a year ago it did this and several people exploited it to get an early lead. However Zen sometimes managed to win anyway by dying but squeezing cut outside groups.
Can we say, in human terms, that the cut is bad early at this point?

Top
 Profile  
 
Offline
 Post subject: Re: DeepZenGo thread
Post #70 Posted: Tue Nov 21, 2017 11:01 am 
Judan

Posts: 6725
Location: Cambridge, UK
Liked others: 436
Was liked: 3719
Rank: UK 4 dan
KGS: Uberdude 4d
OGS: Uberdude 7d
Well, I'm not privy to the latest pro analysis, but before bots started playing the cut my understanding was the cut directly was bad because it could be captured with the micro-Chinese stone one being one line closer than in the mini-Chinese. But if you shoulder hit first then the cut becomes possible: Lee Sedol did this in a game gogameguru commented, but no one else seemed to try that. Then after bots did the cut and died (but with lots of liberties) and then cut and thrashed around on the outside 2 humans tried it in 2017: Ke Jie but he lost: http://ps.waltheri.net/database/game/75078/, Yun Junsang beat Mi Yuting though http://ps.waltheri.net/database/game/75218/. In fact Yoda Norimoto cut in 2012 and died but leaving a long ko.


This post by Uberdude was liked by: Shenoute
Top
 Profile  
 
Offline
 Post subject: Re: DeepZenGo thread
Post #71 Posted: Tue Nov 21, 2017 12:14 pm 
Lives in gote

Posts: 474
Liked others: 120
Was liked: 157
Rank: igs 4d+
A game played today. I like the way black played on the left, the combination of :b7: :b11: and :b13: has a nice post-Alphago flavor (early contact plays and keeping any potential in check). Black :b57:, :b79: and 117 were also nice moves to see. Starting with black 177, there are two kos going on and I was curious to see how Deepzen would handle that since, if I remember well, in a couple of games I posted earlier it seemed that he tried to play both, being the only one playing ko threats while the human kept taking the other ko as his threat.



Attachments:
DEEPZEN(B)-ManTou(P)_20171122 (Tygem).sgf [2.26 KiB]
Downloaded 912 times
Top
 Profile  
 
Offline
 Post subject: Re: DeepZenGo thread
Post #72 Posted: Wed Nov 22, 2017 4:07 am 
Lives with ko

Posts: 136
Liked others: 47
Was liked: 21
Rank: KGS 6 dan
Uberdude wrote:
Well, I'm not privy to the latest pro analysis, but before bots started playing the cut my understanding was the cut directly was bad because it could be captured with the micro-Chinese stone one being one line closer than in the mini-Chinese. But if you shoulder hit first then the cut becomes possible: Lee Sedol did this in a game gogameguru commented, but no one else seemed to try that. Then after bots did the cut and died (but with lots of liberties) and then cut and thrashed around on the outside 2 humans tried it in 2017: Ke Jie but he lost: http://ps.waltheri.net/database/game/75078/, Yun Junsang beat Mi Yuting though http://ps.waltheri.net/database/game/75218/. In fact Yoda Norimoto cut in 2012 and died but leaving a long ko.
Makes me wonder. If Ke jie played it and Yun Junsang beat Mi Yuting in mid 2017, then it probably is a good sacrifice. I tried Mi Yuting's game in Zen7 and Zen thinks W is leading 60% on move 53.

Top
 Profile  
 
Offline
 Post subject: Re: DeepZenGo thread
Post #73 Posted: Wed Nov 22, 2017 5:47 am 
Lives with ko

Posts: 136
Liked others: 47
Was liked: 21
Rank: KGS 6 dan
So I decided to analyze this particular position and, interestingly, commercial version would deal with it differently. I'd say more successfully. I asked all the moves that were boggling me. In all the variations white is better.



Attachments:
Most recent bug analyzed.sgf [1.28 KiB]
Downloaded 827 times
Top
 Profile  
 
Offline
 Post subject: Re: DeepZenGo thread
Post #74 Posted: Sun Dec 03, 2017 2:57 pm 
Lives in gote

Posts: 474
Liked others: 120
Was liked: 157
Rank: igs 4d+
A tengen/mirror go game! Interesting to see that Zen went for a contact play on the tengen stone very early in the game and is the one ending with a large territory in the center. I'm not sure if b's strategy of discarding his three stones in the center to disrupt w's position at the top really paid off considering the situation after move 54 and the aji at H19.



Attachments:
DEEPZEN(B)-Sanli(P)_20171204 (Tygem).sgf [1.61 KiB]
Downloaded 646 times
Top
 Profile  
 
Offline
 Post subject: Re: DeepZenGo thread
Post #75 Posted: Sun Dec 03, 2017 3:42 pm 
Judan

Posts: 6725
Location: Cambridge, UK
Liked others: 436
Was liked: 3719
Rank: UK 4 dan
KGS: Uberdude 4d
OGS: Uberdude 7d
Yeah, I really don't like black sacrificing the centre cutting stones even though top is big profit (but not super big given aji). The previous 5 centre stones become junk and future moves lose so much potential severity from cutting things (and didn't even get f11 peep first). If you are Lee Changho and read all the endgame to your half point win then exceptions apply.

Top
 Profile  
 
Offline
 Post subject: Re: DeepZenGo thread
Post #76 Posted: Wed Feb 28, 2018 6:57 am 
Judan

Posts: 6725
Location: Cambridge, UK
Liked others: 436
Was liked: 3719
Rank: UK 4 dan
KGS: Uberdude 4d
OGS: Uberdude 7d
Zen continues to beat up Japanese pros, but I noticed today wise old Yoda Norimoto scored a rare win by resign (and a loss by 3.5) with a cute little endgame reduction tesuji that Zen missed. Not sure who would have won if Zen avoided this: some moves ago Zen had cut at 2-3 at top left: if black answers this seems to achieve nothing, but if black ignores does it stop a 1st line attachment tesuji on top? Meanwhile Yoda played some gote moves to setup up the tesuji.
Attachment:
yoda zen.png
yoda zen.png [ 362.74 KiB | Viewed 8255 times ]


This post by Uberdude was liked by: Gomoto
Top
 Profile  
 
Display posts from previous:  Sort by  
Post new topic Reply to topic  [ 76 posts ]  Go to page Previous  1, 2, 3, 4

All times are UTC - 8 hours [ DST ]


Who is online

Users browsing this forum: Majestic-12 [Bot] and 1 guest


You cannot post new topics in this forum
You cannot reply to topics in this forum
You cannot edit your posts in this forum
You cannot delete your posts in this forum
You cannot post attachments in this forum

Search for:
Jump to:  
Powered by phpBB © 2000, 2002, 2005, 2007 phpBB Group