| Life In 19x19 http://prod.lifein19x19.com/ |
|
| Fun with sanrensei http://prod.lifein19x19.com/viewtopic.php?f=15&t=16992 |
Page 1 of 2 |
| Author: | Gomoto [ Mon Oct 21, 2019 11:28 am ] |
| Post subject: | Fun with sanrensei |
From time to time somebody tries to play sanrensei against me. Poor souls, they should know better  In other words, I feel a sudden surge of power if my opponents do so and most of the time ... But this move beats them all. This is why I love go. The ultimate bully move. I had a good laugh when it showed up during the analyzing of my last game. I am looking forward to applying it versus my next victims  |
|
| Author: | Gomoto [ Mon Oct 21, 2019 11:30 am ] |
| Post subject: | Re: Fun with sanrensei |
| Author: | Uberdude [ Mon Oct 21, 2019 2:51 pm ] |
| Post subject: | Re: Fun with sanrensei |
One of my favourite pro games, Cho vs Takemiya: https://online-go.com/game/425045 |
|
| Author: | xela [ Mon Oct 21, 2019 3:02 pm ] |
| Post subject: | Re: Fun with sanrensei |
Quote: White's winrate: 64,9% Don't get too excited. KataGo translates this into a score of W+3.8. Nice to have a bit of an edge, but at our amateur level of play it's insignificant compared to the mistakes we're going to make later. |
|
| Author: | Bill Spight [ Mon Oct 21, 2019 3:13 pm ] |
| Post subject: | Re: Fun with sanrensei |
Uberdude wrote: One of my favourite pro games, Cho vs Takemiya: https://online-go.com/game/425045 Interesting.  What's the date? Thanks.
|
|
| Author: | Uberdude [ Mon Oct 21, 2019 3:26 pm ] |
| Post subject: | Re: Fun with sanrensei |
Bill Spight wrote: Uberdude wrote: One of my favourite pro games, Cho vs Takemiya: https://online-go.com/game/425045 Interesting. What's the date? Thanks.Date: 9 October 1996 Event: 21st Japanese Meijin |
|
| Author: | Gomoto [ Mon Oct 21, 2019 3:58 pm ] |
| Post subject: | Re: Fun with sanrensei |
KataGo W +3.5 is good enough for me I have some experience playing against sanrensei as well. And I love it. A move like this has an impact on my opponent that is worth more to me, than if I win the game in the end indeed. I enjoy go for the fun of it as well.
|
|
| Author: | Bill Spight [ Mon Oct 21, 2019 5:24 pm ] |
| Post subject: | Re: Fun with sanrensei |
xela wrote: Quote: White's winrate: 64,9% Don't get too excited. KataGo translates this into a score of W+3.8. Nice to have a bit of an edge, but at our amateur level of play it's insignificant compared to the mistakes we're going to make later. Don't pooh-pooh small errors in points at the opening stage. A 2 pt. mistake in the opening is consistent with a rank difference of more than 9 stones. Sure, it's not a big deal if otherwise one's opening is perfect, but it isn't.
|
|
| Author: | Uberdude [ Mon Oct 21, 2019 9:32 pm ] |
| Post subject: | Re: Fun with sanrensei |
To help calibrate your sense of how big opening mistakes are in winrate and points it is instructive to look at mistakes which you already know and stand out to you. For example, I think even a DDK who has read some books on the opening, and most SDKs can tell that this black move is blocking from the wrong side according to our old theory as it doesn't coordinate with the san ren sei stone: However, KataGo rates this 9 as 42.8% for black and and 2.0 points lead for white, whilst blocking on the 'correct' right side is actually worse at 40.0% and 3 points for white. So perhaps not the best example as it's an instance of old theory being wrong. LZ 247 actually thinks it's a less bad mistake, losing less than 1%, and LZ normally gives bigger winrate changes than KataGo in my small experience so here is an example of their different style. The bots distaste of the old logic of making the san ren sei side stone work with the wall can be understood in the context of them not liking moces on the side, preferring the local efficiency of separating the approach stone from the corner than making some too lose side moyo. So how about a move bots agree with us, such as this misclick which was meant to be classic joseki at a: KataGo rates a at 42.6% for black and white leading 2.0 points, a slight mistake vs best move at 3-3 top left at 44.0% and white 1.5 points, so it lost 1.4% and 0.5 points. The misclick move is 36.1% and white 4.5 points, so compared to the joseki move loses 6.4% and 2.5 points. So if dumb misclicks "only" lose 2.5 points compared to what we thought was correct play, a 3.8 point lead after a few opening moves is similarly good as your opponent being obviously dumb like this. But it's also true that even dan players can make 30+ point mistakes in the middlegame, so avoiding those is more important (my blunder in the recent British championship game was a -57 points according to KataGo, though that was unusually bad). |
|
| Author: | Bill Spight [ Tue Oct 22, 2019 12:39 am ] |
| Post subject: | Re: Fun with sanrensei |
The Elf commentaries tell a somewhat different winrate store than KataGo does. From GoGoD 1994-01-20i, Liu Jun (W) - Yu Keqiang Elf thinks that  loses 3½% to par, within the margin of error, I believe. So far, I don't think I have seen Elf give it a loss of more than 5% (probably a minor error) in any context. (Edit: But see below.  ) Similarly, it thinks that  loses 4½% to par, maybe an error, maybe not. Elf prefers  giving it a winrate estimate of 40.9% (54.4k playouts). By comparison it gives a a winrate estimate of 39.3% (4.7k playouts) for a loss of 1½%. IMHO, because of so few playouts, a should inherit its winrate estimate from Elf's choice of White's reply at 9, i.e., 39.7% (34.7k playouts), for a loss of only 1%. In any event the winrate difference is well within Elf's margin of error. Even though Elf prefers the block at the difference is not enough to say that the block at a is a mistake. In a casual game the surprise factor of may give a psychological advantage.  Elf's mainline variation for is pretty interesting.  (41%, 16.9k playouts) should have some shock value, as well.
|
|
| Author: | Bill Spight [ Tue Oct 22, 2019 12:55 am ] |
| Post subject: | Re: Fun with sanrensei |
Uberdude wrote: So if dumb misclicks "only" lose 2.5 points compared to what we thought was correct play, a 3.8 point lead after a few opening moves is similarly good as your opponent being obviously dumb like this. Well, a misclick is a misclick. However, if Black made a practice of losing 2.5 pts. per move in the opening he'd probably be somewhere in the SDKs. Amateur dans can easily lose 10 - 20 pts. to par in the opening, IMO. Edit: A loss of only 3.8 pts. in 15 moves per player is actually fairly strong play for an amateur. And it's actually somewhat less of a loss, since KataGo, along with most other bots, thinks that White starts with an advantage.
|
|
| Author: | Gomoto [ Tue Oct 22, 2019 5:13 am ] |
| Post subject: | Re: Fun with sanrensei |
I was already quite interested in the 15% difference in evaluation of several middle game positions by KataGo and LZ before this thread. That is quite a margin. I will have to watch some games between Kata and LZ when I have some time and dedication at hand. |
|
| Author: | Bill Spight [ Tue Oct 22, 2019 8:00 am ] |
| Post subject: | Re: Fun with sanrensei |
Sanrensei vs. mukai komoku Ishida Yoshio (W) - Takemiya Masaki, 1976-07-01a, Honinbo final, game 3 The winrate estimate for a play depends in part on other available plays. Elf thinks that Black should approach one of the White corners. Not doing so reduces the winrate estimate for .Later  loses 8% by comarison with a. With the Black moyo on the right side Elf prefers to block at  instead of 24, but only by 2%.
|
|
| Author: | xela [ Thu Oct 24, 2019 3:53 pm ] |
| Post subject: | Re: Fun with sanrensei |
Bill Spight wrote: Don't pooh-pooh small errors in points at the opening stage. A 2 pt. mistake in the opening is consistent with a rank difference of more than 9 stones. Sure, it's not a big deal if otherwise one's opening is perfect, but it isn't. :) Interesting stat. Where did you get it from? I'm thinking that a "2 pt mistake" described as such in a human commentary and a "2 pt mistake" comparing the human move with KataGo's suggestion may not be quite the same thing. Uberdude wrote: One of my favourite pro games, Cho vs Takemiya: https://online-go.com/game/425045 Black makes three "2 pt mistakes" in a row at moves 7-11 according to KataGo. White is also making "mistakes" so the game stays close (black 2-3 points behind, if we believe the AI) for a while. But then Takemiya's play on tengen is another "mistake" (score goes from -3.7 to -5.4; winrate from 40% to 36% based on 10k playouts before and after -- but white's reply was a blunder of equal magnitude). Oh, and thanks Uberdude, I enjoyed going through that game record :-) |
|
| Author: | Bill Spight [ Thu Oct 24, 2019 5:37 pm ] |
| Post subject: | Re: Fun with sanrensei |
xela wrote: Bill Spight wrote: Don't pooh-pooh small errors in points at the opening stage. A 2 pt. mistake in the opening is consistent with a rank difference of more than 9 stones. Sure, it's not a big deal if otherwise one's opening is perfect, but it isn't. Interesting stat. Where did you get it from? Back in the 1970s statistical analysis of Japanese pro-pro handicap games suggested that, up to 9 stones, each handicap stone was worth a little less than 14 pts. The relationship was surprisingly linear. It was on that basis that I predicted that Japanese komi would increase to 6½ pts. by the year 2000. Almost! OC, each stone corresponds to a one amateur rank difference. Now if a player consistently loses 2 pts. per move in the opening, that does not mean that he will average a loss of 2 pts. per move throughout the game. So let's estimate his average loss per move at 1 pt., so he loses around 120 pts. plus per game. And that is approximately equal to the average loss from a 9 rank difference. I misremembered, thinking that the loss from a 9 rank difference was less. OC, for most players their losses are not unimodal, because they have large average losses in the middle game. OC, they are important, and reducing them is significant. But that actually means that a person who loses 120 points overall actually loses less than an average of 2 pts. per move in the opening.
|
|
| Author: | gennan [ Fri Oct 25, 2019 2:47 am ] |
| Post subject: | Re: Fun with sanrensei |
Bill Spight wrote: Back in the 1970s statistical analysis of Japanese pro-pro handicap games suggested that, up to 9 stones, each handicap stone was worth a little less than 14 pts. The relationship was surprisingly linear. It was on that basis that I predicted that Japanese komi would increase to 6½ pts. by the year 2000. Almost! OC, each stone corresponds to a one amateur rank difference.I just did a quick test with KataGo to estimate the actual value of handicap stones in terms of points. I also listed an idealized value using 14 points / stone. Code: handicap actual idealized difference 1 9 7 +2 2 28 21 +7 3 42 35 +7 4 55 49 +6 5 59 63 -4 6 70 77 -7 7 83 91 -8 8 101 105 -4 9 120 119 +1 So indeed it's fairly linear and close to 14 points / stone (within half a stone). It would be interesting to find a simple and symmetric handicap stone placement scheme that would be actually linear with 14 points / stone. |
|
| Author: | Uberdude [ Fri Oct 25, 2019 3:40 am ] |
| Post subject: | Re: Fun with sanrensei |
Interesting how KataGo thinks tengen when going from 4 to 5 stones isn't very useful at just +4 points ("just one more stone to kill" ), but when going from 8 to 9 is +19. But how much has KataGo been trained on positions as lopsided as an initial handicap so that it can make reliable judgements? Or do we just trust the magic extrapolation abilities of neural networks?
|
|
| Author: | gennan [ Fri Oct 25, 2019 4:55 am ] |
| Post subject: | Re: Fun with sanrensei |
Uberdude wrote: Interesting how KataGo thinks tengen when going from 4 to 5 stones isn't very useful at just +4 points ("just one more stone to kill" ), but when going from 8 to 9 is +19. But how much has KataGo been trained on positions as lopsided as an initial handicap so that it can make reliable judgements? Or do we just trust the magic extrapolation abilities of neural networks?I think it reflects "first corners, then sides, then center". The 5th handicap stone at tengen is premature. It's too far from black's other stones to create a good synergy. With 7 stones, there is synergy with the side hoshi and with 9 handicap, tengen is very nicely located and maximizing the synergy of all black's stones. I don't know how accurate KataGo's evaluation is in these positions, but it is fairly consistent with the pro assesment and my feeling of the relative value of the handicap stones. It "feels" right to me. |
|
| Author: | xela [ Fri Oct 25, 2019 5:08 am ] |
| Post subject: | Re: Fun with sanrensei |
Uberdude wrote: Or do we just trust the magic extrapolation abilities of neural networks? No we don't :-) But it's fun to read the tea leaves. Here's a couple of interesting things. First, the winrates (my scores are only slightly different from gennan's: this could be the random number see or a different number of playouts. I'm using 15,000). Code: handicap score winrate 1 9.0 74.6% 2 28.7 96.9% 3 44.0 98.8% 4 56.8 99.2% 5 59.8 99.1% So for more than 2 stones, the winrate difference for each extra stone is too small to measure accurately. Second, the above numbers (and I guess gennan's too) are obtained by putting the handicap stones on the board, set komi to zero and ask KataGo to evaluate the position. But if you try different choices of komi, then it explores different moves, and this affects the results. On 2 stones with 29 points komi, it puts black 7 points behind, winrate 31%. But 2 stones with 22 points komi gives black only 0.5 point behind, 48% winrate. On 4 and 5 stones, the most balanced komi is 53 points for 4 stones and 63 points for 5 stones. So by this measure, the 5th stone might be worth a bit less, but the difference isn't so drastic. Is either measure more accurate (or less meaningless) than the other? |
|
| Author: | xela [ Fri Oct 25, 2019 5:18 am ] |
| Post subject: | Re: Fun with sanrensei |
Oh, for completeness: Code: handicap komi for balance difference from previous row 1 7 N/A 2 22 15 3 37 15 4 53 16 5 63 10 Yes, if you run KataGo on an empty board with no komi, it tells you black is 9 points ahead, but if you give white 9 points komi, you don't get an equal position, you get a position with black just over 3 points behind. That's more or less what we'd expect if fair komi is 7. |
|
| Page 1 of 2 | All times are UTC - 8 hours [ DST ] |
| Powered by phpBB © 2000, 2002, 2005, 2007 phpBB Group http://www.phpbb.com/ |
|