Life In 19x19 http://www.lifein19x19.com/ |
|
Opening problems for AI: Problem 43 http://www.lifein19x19.com/viewtopic.php?f=15&t=17718 |
Page 1 of 1 |
Author: | Bill Spight [ Mon Aug 10, 2020 7:33 pm ] |
Post subject: | Opening problems for AI: Problem 43 |
Only move problem. White to play. Enjoy! ![]() |
Author: | Uberdude [ Mon Aug 10, 2020 11:38 pm ] |
Post subject: | Re: Opening problems for AI: Problem 43 |
Author: | Knotwilg [ Tue Aug 11, 2020 1:54 am ] |
Post subject: | Re: Opening problems for AI: Problem 43 |
Two possibilities: |
Author: | Joaz Banbeck [ Tue Aug 11, 2020 5:46 am ] |
Post subject: | Re: Opening problems for AI: Problem 43 |
This seems to preserve white's eye space as well as any move. |
Author: | Kirby [ Tue Aug 11, 2020 7:48 am ] |
Post subject: | Re: Opening problems for AI: Problem 43 |
I'd just play here: or maybe here: |
Author: | Harleqin [ Tue Aug 11, 2020 9:33 am ] |
Post subject: | Re: Opening problems for AI: Problem 43 |
Author: | Joaz Banbeck [ Tue Aug 11, 2020 12:34 pm ] |
Post subject: | Re: Opening problems for AI: Problem 43 |
I'm highly skeptical of tenukiing to a corner. With Bill's #42 still in mind, I'm seeing honte moves in my sleep. But I don't like playing the corner either. My gut says that a five point move in gote close to black's strength can't be best. Perhaps there is a possibility in between. |
Author: | Uberdude [ Tue Aug 11, 2020 2:23 pm ] |
Post subject: | Re: Opening problems for AI: Problem 43 |
Joaz Banbeck wrote: Not as well as this move, which I can only see advantages and no disadvantages over your move, except perhaps making the opponent lose concentration from the shock of such a bizarre move! How can you go from so right on the last problem to so wrong here?! |
Author: | ez4u [ Tue Aug 11, 2020 3:40 pm ] |
Post subject: | Re: Opening problems for AI: Problem 43 |
Bill Spight wrote: Only move problem. White to play... Enjoy! ![]() Huh? But you said... ![]() Bill Spight wrote: "Tenuki is always an option!" ![]() |
Author: | Bill Spight [ Tue Aug 11, 2020 4:11 pm ] |
Post subject: | Re: Opening problems for AI: Problem 43 |
ez4u wrote: Bill Spight wrote: Only move problem. White to play... Enjoy! ![]() Huh? But you said... ![]() Bill Spight wrote: "Tenuki is always an option!" ![]() ![]() |
Author: | Gomoto [ Tue Aug 11, 2020 4:17 pm ] |
Post subject: | Re: Opening problems for AI: Problem 43 |
A hideous insidious riddle it is I thought about B3 (mild) spoiler |
Author: | Bill Spight [ Wed Aug 12, 2020 9:15 pm ] |
Post subject: | Re: Opening problems for AI: Problem 43 |
Author: | yoyoma [ Wed Aug 12, 2020 10:10 pm ] |
Post subject: | Re: Opening problems for AI: Problem 43 |
Code: a - 4-4 upper left : White 49.5% -0.1pts (Kata's choice) b - solid connect : White 47.1% -0.2pts (Elf's "only move") c - 4-4 upper right : White 46.8% -0.2pts d - 5-4 upper right : White 41.5% -0.9pts (Game move) e - shoulder hit : White 43.6% -0.7pts f - block corner : White 36.1% -1.7pts g - descend : White 4.3% -10.2pts Kata is not so insistent about connecting. It's a good move, but so is taking the corners. If White takes a corner, Kata also thinks Black can choose taking a corner or playing the forcing push. |
Author: | Bill Spight [ Wed Aug 12, 2020 10:28 pm ] |
Post subject: | Re: Opening problems for AI: Problem 43 |
yoyoma wrote: Code: a - 4-4 upper left : White 49.5% -0.1pts (Kata's choice) b - solid connect : White 47.1% -0.2pts (Elf's "only move") c - 4-4 upper right : White 46.8% -0.2pts d - 5-4 upper right : White 41.5% -0.9pts (Game move) e - shoulder hit : White 43.6% -0.7pts f - block corner : White 36.1% -1.7pts g - descend : White 4.3% -10.2pts Kata is not so insistent about connecting. It's a good move, but so is taking the corners. If White takes a corner, Kata also thinks Black can choose taking a corner or playing the forcing push. This is great! ![]() Elf apparently developed a blind spot that kept it from exploring alternatives. It is extremely rare for it to like a play so much that it gives it over 100k rollouts. How many rollouts does KataGo give these different options? |
Author: | Uberdude [ Thu Aug 13, 2020 1:11 am ] |
Post subject: | Re: Opening problems for AI: Problem 43 |
I've seen Elf (v2 I think) have a blindspot and some % winrate change when shown the shockingly unusual move 3 of a 3-4 point instead of a 4-4. Other bots are less blinkered and more reliable for analysis IMO. |
Author: | Bill Spight [ Thu Aug 13, 2020 6:29 am ] |
Post subject: | Re: Opening problems for AI: Problem 43 |
Uberdude wrote: I've seen Elf (v2 I think) have a blindspot and some % winrate change when shown the shockingly unusual move 3 of a 3-4 point instead of a 4-4. Other bots are less blinkered and more reliable for analysis IMO. Well, the bots are at present good tools to assist a human analyst, but I would not rely upon any computer program alone for analysis. They are not trained to be analysts. Now, lightvector has used human games in training KataGo to make it better for analysis, but the main impetus for bot development is to increase its overall playing strength. The same seems to be true for chess engines, some 25 years or so since the first superhuman engines appeared. Chess engines still have well known blind spots. ![]() Ordinarily I would not be surprised that KataGo or any other bot would prefer a different play than Elf. From what little information I have, a concordance rate between bots of around 80% is not unexpected. I do expect bots to agree about sufficiently bad human errors, which is the case here. KataGo figures a winrate loss for Sansa's play at 8%, about half the loss that Elf figures, but that's in line with how differently they estimate winrates. I am however surprised that, having spent at least 143.8k rollouts on this position, Elf did not pick up KataGo's top choice and give it at least 1500 rollouts. |
Author: | ez4u [ Thu Aug 13, 2020 6:37 am ] |
Post subject: | Re: Opening problems for AI: Problem 43 |
Bill Spight wrote: yoyoma wrote: Code: a - 4-4 upper left : White 49.5% -0.1pts (Kata's choice) b - solid connect : White 47.1% -0.2pts (Elf's "only move") c - 4-4 upper right : White 46.8% -0.2pts d - 5-4 upper right : White 41.5% -0.9pts (Game move) e - shoulder hit : White 43.6% -0.7pts f - block corner : White 36.1% -1.7pts g - descend : White 4.3% -10.2pts Kata is not so insistent about connecting. It's a good move, but so is taking the corners. If White takes a corner, Kata also thinks Black can choose taking a corner or playing the forcing push. This is great! ![]() Elf apparently developed a blind spot that kept it from exploring alternatives. It is extremely rare for it to like a play so much that it gives it over 100k rollouts. How many rollouts does KataGo give these different options? First, here is katago's policy. Elf's preferred play is high but not as high as the upper left. Of course the PO's depend on how long we search. ![]() But after 2.8 million... (note my komi was set to 6.5 points and therefore the winrates are quite different) |
Author: | Bill Spight [ Thu Aug 13, 2020 7:35 am ] |
Post subject: | Re: Opening problems for AI: Problem 43 |
ez4u wrote: First, here is katago's policy. Elf's preferred play is high but not as high as the upper left. Of course the PO's depend on how long we search. ![]() But after 2.8 million... (note my komi was set to 6.5 points and therefore the winrates are quite different) This is great, Dave! Many thanks. ![]() ![]() ![]() Two comments. First, with only 521 playouts, KataGo's evaluation of Sansa's 5-4 move in the top right has a large margin of error. Elf does better in this regard, by inheriting the evaluation from its best reply to that move. I expect that yoyoma does even better, by making the play itself to get its evaluation with several thousand rollouts. Second, even though I have no experience with KataGo, I think that the 3% winrate difference between KataGo's choice and Elf's, with 1.6M and 147K rollouts, respectively, puts Elf's choice in the doubtful category. The fact that Elf concentrates on one play to the virtual exclusion of others indicates that it likes that play. However, that does not mean that other bots will like it much. Therefore, this will be the last "only move" problem. |
Author: | Gomoto [ Thu Aug 13, 2020 7:39 am ] |
Post subject: | Re: Opening problems for AI: Problem 43 |
Quote: Therefore, this will be the last "only move" problem. The only move now ![]() |
Author: | dhu163 [ Mon Dec 26, 2022 9:10 am ] |
Post subject: | Re: Opening problems for AI: Problem 43 |
Page 1 of 1 | All times are UTC - 8 hours [ DST ] |
Powered by phpBB © 2000, 2002, 2005, 2007 phpBB Group http://www.phpbb.com/ |