Life In 19x19 http://lifein19x19.com/ |
|
Change in joseki choice http://lifein19x19.com/viewtopic.php?f=12&t=16792 |
Page 1 of 1 |
Author: | Hades12 [ Wed Jul 24, 2019 3:10 pm ] |
Post subject: | Change in joseki choice |
Curious about the resurfacing of the joseki in diagram 2. The modern variation was diagram 1 for a long time. Did diagram 2 come back into fashion due to the AI style of invading everywhere? I assume in the second variation there is less risk with the three marked stones. The invasion point a 'a' is easier to handle in the second diagram than in the first diagram. |
Author: | Uberdude [ Wed Jul 24, 2019 3:33 pm ] |
Post subject: | Re: Change in joseki choice |
Early versions of AlphaGo tended to prefer the hanging connection, but this changed to a preference for the solid in later and stronger versions. LZ is similar. The main reason seems to be because the hanging connection is not sente, white doesn't need to make the one point jump (and to do so is slack in early opening) because black's attach at r5 is not so severe with hanging. Indeed with the solid the followup if ignored is not simply to extend at r6 for gote thickness as was normal pre-AI, but to severely counter hane at s6, a move Go Seigen also recommended. Also AI tend to think solid connect and then tenuki (usually not a total tenuki but not the joseki extension) is ok, and better than tenuki one move earlier, which is different to the previous theory that it's better to tenuki before connecting as that makes you heavy, but actually it's quite hard to kill the 3 stones whereas if you let white cut white gets strong shape. |
Author: | Bill Spight [ Wed Jul 24, 2019 4:30 pm ] |
Post subject: | Re: Change in joseki choice |
Uberdude wrote: AI tend to think solid connect and then tenuki (usually not a total tenuki but not the joseki extension) is ok, and better than tenuki one move earlier, which is different to the previous theory that it's better to tenuki before connecting as that makes you heavy, but actually it's quite hard to kill the 3 stones whereas if you let white cut white gets strong shape. Also, when the extension is made on the side, the old joseki typically had a 3d line extension. The modern extension on the 4th line instead, as in the diagram, raises questions about the claim that modern AI are territory players. |
Author: | ez4u [ Wed Jul 24, 2019 7:51 pm ] |
Post subject: | Re: Change in joseki choice |
This topic is laboring under some misapprehensions with regard to Go history. Let's check the facts (at least as they appear in my database! ). I have 5,159 games with the position below (from even games played through 2016 and ignoring color). As noted in the thread, the hanging connection at "a" is more frequently seen (1,049 games, B winning 46%) than the solid connection with "B" (704 games, B winning 49%). However, the real "modern variation" was not "a", it was the approach move at "c" (1,922 games, B winning 52%). Beyond that if we look closely at the joseki discussed in the thread... This is the most frequent actual continuation from the direct hanging connection (359 games, B winning 51%). The simple extension with White playing the large-knight's move appears 243 times with B winning only 40%. That unsatisfactory performance is what led Black to switch to approaching from the other side at "c" in the diagram above. The solid connection with the high extension shown below actually appeared more often than any single variation of the immediate hanging connection (438 games, B winning 50%). Note that White often played tenuki here. If White did approach it was at "a" (190 games) rather than "b" (3 games). And sorry Bill, against the star point Black played the low extension in only 98 games, winning 44% of them. |
Author: | Kirby [ Wed Jul 24, 2019 8:32 pm ] |
Post subject: | Re: Change in joseki choice |
ez4u wrote: This topic is laboring under some misapprehensions with regard to Go history. Let's check the facts (at least as they appear in my database! ). I have 5,159 games with the position below (from even games played through 2016 and ignoring color). As noted in the thread, the hanging connection at "a" is more frequently seen (1,049 games, B winning 46%) than the solid connection with "B" (704 games, B winning 49%). However, the real "modern variation" was not "a", it was the approach move at "c" (1,922 games, B winning 52%). FWIW, I tried this out on whatever version of Leela Zero I have on my computer - board position being the same, except with two 4-4 black stones on the right side. When black approaches at 'c', LZ wants to just cut at 'b': If I go ahead and play out further along the popular sequence: LZ wants to tenuki now and play 3-3 Go figure :-p |
Author: | Bill Spight [ Wed Jul 24, 2019 10:15 pm ] |
Post subject: | Re: Change in joseki choice |
ez4u wrote: This topic is laboring under some misapprehensions with regard to Go history. Let's check the facts (at least as they appear in my database! ). {snip} The solid connection with the high extension shown below actually appeared more often than any single variation of the immediate hanging connection (438 games, B winning 50%). Note that White often played tenuki here. If White did approach it was at "a" (190 games) rather than "b" (3 games). And sorry Bill, against the star point Black played the low extension in only 98 games, winning 44% of them. Sorry, Dave, I was talking more generally, not simply when there was a 4-4 stone in the bottom left corner. I considered adding the following diagram and saying a bit more, but decided not to take the time. Top bots will also play the high extension, , in other situations, such as when White has an enclosure in the bottom left corner. The traditional joseki is to make a low extension; the modern, AI, joseki is to tenuki. Right, Dave? |
Page 1 of 1 | All times are UTC - 8 hours [ DST ] |
Powered by phpBB © 2000, 2002, 2005, 2007 phpBB Group http://www.phpbb.com/ |