Predict the winner of World Go Championship 2017
-
pookpooi
- Lives in sente
- Posts: 727
- Joined: Sat Aug 21, 2010 12:26 pm
- GD Posts: 10
- Has thanked: 44 times
- Been thanked: 218 times
Re: Predict the winner of World Go Championship 2017
Preliminary Draw
First day: DeepZenGo vs. Mi Yuting, Park Jungwhan vs. Iyama Yuta
Second Day: Park Jungwhan vs. DeepZenGo, Mi Yuting vs. Iyama Yura
Third Day: Iyama Yuta vs. DeepZenGo, Park Jungwhan vs. Mi Yuting
First day: DeepZenGo vs. Mi Yuting, Park Jungwhan vs. Iyama Yuta
Second Day: Park Jungwhan vs. DeepZenGo, Mi Yuting vs. Iyama Yura
Third Day: Iyama Yuta vs. DeepZenGo, Park Jungwhan vs. Mi Yuting
-
pookpooi
- Lives in sente
- Posts: 727
- Joined: Sat Aug 21, 2010 12:26 pm
- GD Posts: 10
- Has thanked: 44 times
- Been thanked: 218 times
Re: Predict the winner of World Go Championship 2017
Park Jungwhan is 2 hours late
Edit: He's not late, but just in time for the match (15 minutes) but it looks like everyone is waiting for him to arrive
Edit: He's not late, but just in time for the match (15 minutes) but it looks like everyone is waiting for him to arrive
-
pookpooi
- Lives in sente
- Posts: 727
- Joined: Sat Aug 21, 2010 12:26 pm
- GD Posts: 10
- Has thanked: 44 times
- Been thanked: 218 times
Re: Predict the winner of World Go Championship 2017
If you watch through Nico video, you can see winrate provide by DeepZenGo for both game (though not at the same time)
And sometimes they'll show heat map too (but like once every 30 minutes or so)
And sometimes they'll show heat map too (but like once every 30 minutes or so)
-
pookpooi
- Lives in sente
- Posts: 727
- Joined: Sat Aug 21, 2010 12:26 pm
- GD Posts: 10
- Has thanked: 44 times
- Been thanked: 218 times
Re: Predict the winner of World Go Championship 2017
The same thing happen when it match Cho Chikun last year.
Anyway, Iyama Yuta winrate drop to 30%, but does this mean he still has a chance? If Zen's evaluation can't be trusted.
Anyway, Iyama Yuta winrate drop to 30%, but does this mean he still has a chance? If Zen's evaluation can't be trusted.
-
kimidori
- Dies with sente
- Posts: 112
- Joined: Wed Jun 12, 2013 3:41 am
- Rank: KGS 3d
- GD Posts: 0
- Has thanked: 26 times
- Been thanked: 4 times
Re: Predict the winner of World Go Championship 2017
Indeed this time Zen can be trusted. Park won, and I also think that he started to gain the advantage after the complex fight on the left side/center.
-
Uberdude
- Judan
- Posts: 6727
- Joined: Thu Nov 24, 2011 11:35 am
- Rank: UK 4 dan
- GD Posts: 0
- KGS: Uberdude 4d
- OGS: Uberdude 7d
- Location: Cambridge, UK
- Has thanked: 436 times
- Been thanked: 3718 times
Re: Predict the winner of World Go Championship 2017
I think things started going south for Iyama when he played that high move on the centre left (move 120, below), Park plonked a stone inside and Iyama ignored to bolster the connection of the wall above.
Before that Michael Redmond thought the game was ok/nice for Iyama as he only needed about 30 points from that left side which looked very doable with sente, but then Iyama asked for a lot more and didn't try to kill black's invasion. So I have 2 questions about this sequence. First as Park (black) gave Iyama a few extra points for the right group to make sure his top right group was alive the threat of saving the cutting stone in the centre was a lot smaller, so why did Iyama play so high (one secondary aim of which was presumably threatening to save that stone, as well as helping a bit with the cuts in the wall above and surrounding the left) if it meant Park could easily invade. Secondly why tenuki the invasion and allow him to trash the side if you asked for more? No good answer? Would a more conservative move, such as one-point jump up from the 3-10 side stone (maybe too slow) or a large knight's move (2 spaces left of where he played, feels good to me) not be enough? Perhaps he feared being whittled down from all sides (black has some nice reduction from top left corner, cut the 3 stones in centre, corner slide in lower left) and wouldn't have enough? So he plays a big move asking for more, ignores the invasion to play the thick connection/yose at the the top left (forming an ugly empty triangle) and then defend the cut in the wall and aim to attack the invader on a large scale after it breaks the left side territory.
As part of that attack he pressed the lower side, threatening to take a chunky bottom-left corner with 146 (3 below) if Park is generous and takes care of his invasion group for 147 instead of the most severe move he played of extending into corner (4 below). In that case I can appreciate his high move turns into solid territory above, solid territory below, and weakish black group in middle to tease which is more concrete profit than one big but thin and not-quite-solid large territory getting poked at from all sides (such poking also giving black a bit of additional territory). But with 147 Park challenged Iyama, "go on show me how you build a miai of saving the 2nd line cut or killing the left invasion". I think against his Japanese peers Iyama tends to manage a good result in this kind of complex fighting, but against Park he kind-of failed: yes he got a ko but it was messy and the bottom right wasn't alive yet and black had ko threats, so he got indigestion trying to eat those stones (and his group there had bad shape so chances to die thus heavy ko) and lost the top group in exchange.
What were the thoughts of Japanese/Chinese/Korean commentators about this critical stage of the game? I was pretty sure Park would win by move 143 (end of 3rd diagram).
Before that Michael Redmond thought the game was ok/nice for Iyama as he only needed about 30 points from that left side which looked very doable with sente, but then Iyama asked for a lot more and didn't try to kill black's invasion. So I have 2 questions about this sequence. First as Park (black) gave Iyama a few extra points for the right group to make sure his top right group was alive the threat of saving the cutting stone in the centre was a lot smaller, so why did Iyama play so high (one secondary aim of which was presumably threatening to save that stone, as well as helping a bit with the cuts in the wall above and surrounding the left) if it meant Park could easily invade. Secondly why tenuki the invasion and allow him to trash the side if you asked for more? No good answer? Would a more conservative move, such as one-point jump up from the 3-10 side stone (maybe too slow) or a large knight's move (2 spaces left of where he played, feels good to me) not be enough? Perhaps he feared being whittled down from all sides (black has some nice reduction from top left corner, cut the 3 stones in centre, corner slide in lower left) and wouldn't have enough? So he plays a big move asking for more, ignores the invasion to play the thick connection/yose at the the top left (forming an ugly empty triangle) and then defend the cut in the wall and aim to attack the invader on a large scale after it breaks the left side territory.
As part of that attack he pressed the lower side, threatening to take a chunky bottom-left corner with 146 (3 below) if Park is generous and takes care of his invasion group for 147 instead of the most severe move he played of extending into corner (4 below). In that case I can appreciate his high move turns into solid territory above, solid territory below, and weakish black group in middle to tease which is more concrete profit than one big but thin and not-quite-solid large territory getting poked at from all sides (such poking also giving black a bit of additional territory). But with 147 Park challenged Iyama, "go on show me how you build a miai of saving the 2nd line cut or killing the left invasion". I think against his Japanese peers Iyama tends to manage a good result in this kind of complex fighting, but against Park he kind-of failed: yes he got a ko but it was messy and the bottom right wasn't alive yet and black had ko threats, so he got indigestion trying to eat those stones (and his group there had bad shape so chances to die thus heavy ko) and lost the top group in exchange.
What were the thoughts of Japanese/Chinese/Korean commentators about this critical stage of the game? I was pretty sure Park would win by move 143 (end of 3rd diagram).
Last edited by Uberdude on Tue Mar 21, 2017 6:56 am, edited 5 times in total.
-
Uberdude
- Judan
- Posts: 6727
- Joined: Thu Nov 24, 2011 11:35 am
- Rank: UK 4 dan
- GD Posts: 0
- KGS: Uberdude 4d
- OGS: Uberdude 7d
- Location: Cambridge, UK
- Has thanked: 436 times
- Been thanked: 3718 times
Re: Predict the winner of World Go Championship 2017
kimidori wrote:Mi won instead. Look like Zen's judgement hasn't improved that much since the match with Cho Chikun.
I don't know about that, Mi is much stronger than Cho and it was a very close game. Those dumb moves at the end (some pointless silly sentes, but that atari inside middle left actually lost 1 point) could be because it thought it was winning and would still win despite them:
Hideki Kato on computer-go mailing list wrote:The value network has been trained with Chinese rules and 7.5
pts komi. Using this for Japanese and 6.5, there will be some
error in close games. We knew this issue and thought such
chances would be so small that postponed correcting (not so
easy).
Almost seems like they should turn off the value network and just use monte carlo playouts in the small yose if not doing so means turning a game you are winning by half but think you are winning by 1.5 into a half point loss! (Not saying Zen was winning by half this game).
-
pookpooi
- Lives in sente
- Posts: 727
- Joined: Sat Aug 21, 2010 12:26 pm
- GD Posts: 10
- Has thanked: 44 times
- Been thanked: 218 times
Re: Predict the winner of World Go Championship 2017
I always thought Zen was trained with Japanese game since the team might get Japanese game record database easily. And Zen always play in Japanese rule.
-
pookpooi
- Lives in sente
- Posts: 727
- Joined: Sat Aug 21, 2010 12:26 pm
- GD Posts: 10
- Has thanked: 44 times
- Been thanked: 218 times
Re: Predict the winner of World Go Championship 2017
News report on this part using google translate
Representative Hideki Kato says, "China rules and Japanese rules are almost similar in the calculation of the ground, but in some cases there is room for a difference of 1 st. However, all Go artificial intelligence programs including DeepZenGo are Chinese rules It is made to calculate the formation with.
It is not easy to calculate with Japanese rules. You can force it to apply it, but it can sometimes get twisted if you change the setting. For your information, AlphaGo has tried to solve such a problem and I have heard rumors that it finally resolved. However, to the last it refers to unidentified information. "After all, homework remains in DeepZenGo.
Mi · Witin said, "I was dangerous in the midst of making a mistake, but at the end I can see a bit of some things DeepZenGo needs to improve."
http://nitro15.ldblog.jp/archives/49704509.html
Representative Hideki Kato says, "China rules and Japanese rules are almost similar in the calculation of the ground, but in some cases there is room for a difference of 1 st. However, all Go artificial intelligence programs including DeepZenGo are Chinese rules It is made to calculate the formation with.
It is not easy to calculate with Japanese rules. You can force it to apply it, but it can sometimes get twisted if you change the setting. For your information, AlphaGo has tried to solve such a problem and I have heard rumors that it finally resolved. However, to the last it refers to unidentified information. "After all, homework remains in DeepZenGo.
Mi · Witin said, "I was dangerous in the midst of making a mistake, but at the end I can see a bit of some things DeepZenGo needs to improve."
http://nitro15.ldblog.jp/archives/49704509.html
-
yoyoma
- Lives in gote
- Posts: 653
- Joined: Mon Apr 19, 2010 8:45 pm
- GD Posts: 0
- Location: Austin, Texas, USA
- Has thanked: 54 times
- Been thanked: 213 times
Re: Predict the winner of World Go Championship 2017
Uberdude wrote:kimidori wrote:Mi won instead. Look like Zen's judgement hasn't improved that much since the match with Cho Chikun.
I don't know about that, Mi is much stronger than Cho and it was a very close game. Those dumb moves at the end (some pointless silly sentes, but that atari inside middle left actually lost 1 point) could be because it thought it was winning and would still win despite them:Hideki Kato on computer-go mailing list wrote:The value network has been trained with Chinese rules and 7.5
pts komi. Using this for Japanese and 6.5, there will be some
error in close games. We knew this issue and thought such
chances would be so small that postponed correcting (not so
easy).
Almost seems like they should turn off the value network and just use monte carlo playouts in the small yose if not doing so means turning a game you are winning by half but think you are winning by 1.5 into a half point loss! (Not saying Zen was winning by half this game).
Was there some special situation in the game that made Chinese vs Japanese rules give different results? I didn't see anything unusual, and I thought normally they would give the same result.
-
baduk
- Dies with sente
- Posts: 76
- Joined: Fri Sep 16, 2016 2:37 pm
- GD Posts: 0
- Has thanked: 1 time
- Been thanked: 18 times
Re: Predict the winner of World Go Championship 2017
Yes,chinese rules make a difference more often one may think,for instance the 2015 myly cup finals last game,lee sedol lost by using chinese counting method,if japanese would have been used he would have by 0,5 but lost by 1.5 actually.Till the end lee sedol thought he was winning since he was counting with japanese rules in his mind.I somehow prefer japanese rules since they are more clear to me but its hard to say.