Life In 19x19 http://lifein19x19.com/ |
|
Opening problems for AI: Problem 46 http://lifein19x19.com/viewtopic.php?f=15&t=17728 |
Page 1 of 1 |
Author: | Bill Spight [ Tue Aug 18, 2020 9:37 am ] |
Post subject: | Opening problems for AI: Problem 46 |
White to play. As always, assume komi. Which is better, the extension to a or the pincer at b? Enjoy! |
Author: | Joaz Banbeck [ Tue Aug 18, 2020 10:41 am ] |
Post subject: | Re: Opening problems for AI: Problem 46 |
IMHO, 'a' is the better move The position of makes the difference. Id is captured, then is merely a nusiance. It can mostly be ignored by black, and very possibly be swallowed on a large scale without touching it. If lives, then is the stepping stone for a white attack. With a white stone at G3, white's play at C5 is sente, threatening B3. |
Author: | jlt [ Tue Aug 18, 2020 1:50 pm ] |
Post subject: | Re: Opening problems for AI: Problem 46 |
Author: | Shaddy [ Wed Aug 19, 2020 1:32 am ] |
Post subject: | Re: Opening problems for AI: Problem 46 |
Joaz Banbeck wrote: I agree, but think 2 is gote. |
Author: | Joaz Banbeck [ Wed Aug 19, 2020 5:13 am ] |
Post subject: | Re: Opening problems for AI: Problem 46 |
Shaddy wrote: I extended to C8 because I didn't like this: As fights unfurl in each of the lower corners, black's groups will become stronger. White's middle group will look relatively weak. White has three groups that will need stones, black has two. |
Author: | Gomoto [ Wed Aug 19, 2020 6:07 am ] |
Post subject: | Re: Opening problems for AI: Problem 46 |
Author: | Joaz Banbeck [ Wed Aug 19, 2020 6:52 am ] |
Post subject: | Re: Opening problems for AI: Problem 46 |
[admin] Sorry, folks, we had a leak from an alternate dimension - one in which white had played at C3. TTBOOK, only one post got through. Quick work by your admins has sealed the tear, and we expect no further problems from it. [/admin] |
Author: | jlt [ Wed Aug 19, 2020 7:04 am ] |
Post subject: | Re: Opening problems for AI: Problem 46 |
Gomoto lives in dimension (3,3). |
Author: | Knotwilg [ Wed Aug 19, 2020 9:41 am ] |
Post subject: | Re: Opening problems for AI: Problem 46 |
Author: | Shaddy [ Wed Aug 19, 2020 10:42 am ] |
Post subject: | Re: Opening problems for AI: Problem 46 |
Joaz Banbeck wrote: Shaddy wrote: I extended to C8 because I didn't like this: As fights unfurl in each of the lower corners, black's groups will become stronger. White's middle group will look relatively weak. White has three groups that will need stones, black has two. ok, i don't like this, but i think the problem is 3 not 1. maybe let's play 3 as kosumi from 3-4? |
Author: | Knotwilg [ Wed Aug 19, 2020 12:07 pm ] |
Post subject: | Re: Opening problems for AI: Problem 46 |
@Joaz Surely Black will play at A instead? And so, is a mistake White should simply play and then Black has to decide how to deal with the bottom. attacks the pincer but is already pincered back by the marked stone, which is disposable due to the - having been forcing upon Black. |
Author: | Bill Spight [ Fri Aug 21, 2020 1:10 pm ] |
Post subject: | Re: Opening problems for AI: Problem 46 |
This seems to have been a better problem than I anticipated. It has sparked some very interesting discussion. I am including the GoGoD SGF file for the whole game, because this is the game where Dochi plays a famous endgame tesuji. See B225. |
Author: | Tryss [ Fri Aug 21, 2020 1:51 pm ] |
Post subject: | Re: Opening problems for AI: Problem 46 |
What's interesting is an alternative move by Katago : It has lower policy than a, but about the same winrate/score, and become the "blue move" with enough playouts ( over 100k in my case) |
Author: | Bill Spight [ Fri Aug 21, 2020 3:32 pm ] |
Post subject: | Re: Opening problems for AI: Problem 46 |
Tryss wrote: What's interesting is an alternative move by Katago : It has lower policy than a, but about the same winrate/score, and become the "blue move" with enough playouts ( over 100k in my case) Very interesting! Unlike the pincer, will not become a liability. After White plays , what is KataGo's top choice for ? a? |
Author: | Tryss [ Fri Aug 21, 2020 4:10 pm ] |
Post subject: | Re: Opening problems for AI: Problem 46 |
Black play at a, then white tenuki in the top right corner with P17. The game could look like this : And the position after is interesting : Katago* consider the three moves a, b and c as very close. After 41389 playouts, the results are : - a has 16k visits, 57.5%wr and +0.9pts - b has 10k visits, 57.3%wr and +0.9pts - c has 12k visits, 57.2%wr and +0.9pts *) with the 20b network from the end of april, 7.5 komi and japanese rules |
Author: | lightvector [ Fri Aug 21, 2020 7:20 pm ] |
Post subject: | Re: Opening problems for AI: Problem 46 |
Tryss wrote: What's interesting is an alternative move by Katago : It has lower policy than a, but about the same winrate/score, and become the "blue move" with enough playouts ( over 100k in my case) Interestingly, the 40-block network seems to differ here a little, and prefer's ELF's move more. KataGo seems to have different preferences than ELF in a number of other places in one of Bill's presented lines. It's hard to say whether for sure whether they are better or worse in some cases. Prefers this slightly over "a": Prefers this slightly over "a". Deeper search suggests this doesn't necessarily lead to a better result, but is also interesting and tricky for white to deal with: Some variations for the diagonal: Unlike the above two cases which one might classify as just a difference in preference, in this case, KataGo thinks ELF's "a" for black is a noticeable mistake, and prefers 1-2-3 to take a sente exchange and not let white have the chance to run in the first place. White's lone stone isn't entirely dead, but black has control: White's cut doesn't work, which means that at the expense of strengthening white with a not-great exchange for the outside, black has managed to kill white's two stones in sente, instead of gote. Note that if black were to exchange "a"-"b" below, then white's "c" would work here due to a squeeze and ladder, this is the reason for not playing the peep and why sente is possible if the peep is omitted. |
Author: | Bill Spight [ Fri Aug 21, 2020 9:28 pm ] |
Post subject: | Re: Opening problems for AI: Problem 46 |
Wow! You gotta love those 18th century games. |
Author: | dhu163 [ Mon Dec 26, 2022 9:51 am ] |
Post subject: | Re: Opening problems for AI: Problem 46 |
Page 1 of 1 | All times are UTC - 8 hours [ DST ] |
Powered by phpBB © 2000, 2002, 2005, 2007 phpBB Group http://www.phpbb.com/ |