Life In 19x19 http://www.lifein19x19.com/ |
|
Where would you play? http://www.lifein19x19.com/viewtopic.php?f=15&t=15565 |
Page 1 of 2 |
Author: | Uberdude [ Tue Apr 03, 2018 3:00 am ] |
Post subject: | Where would you play? |
This is a game position with white to play, Komi 6.5, black has 7 captures to white's 3. Last move was Context: |
Author: | Schachus [ Tue Apr 03, 2018 4:23 am ] |
Post subject: | Re: Where would you play? |
I'm not at all a dan player, but I will still try my luck: |
Author: | bernds [ Tue Apr 03, 2018 5:04 am ] |
Post subject: | Re: Where would you play? |
Uberdude wrote: |
Author: | lightvector [ Tue Apr 03, 2018 5:10 am ] |
Post subject: | Re: Where would you play? |
Edit: added one more diagram to the end. |
Author: | Javaness2 [ Tue Apr 03, 2018 5:16 am ] |
Post subject: | Re: Where would you play? |
Author: | Uberdude [ Tue Apr 03, 2018 5:18 am ] |
Post subject: | Re: Where would you play? |
reply to bernds: |
Author: | lightvector [ Tue Apr 03, 2018 5:30 am ] |
Post subject: | Re: Where would you play? |
Oh, in case anyone is interested, I've been training some neural nets recently, and here's what my biggest neural net says on the position (without move history as an input, although I tried giving it history and it doesn't affect the heatmap that much). It's a 12 block residual net with about 8 million parameters trained to predict GoGoD pro games. The raw policy (no search, no value) beats Pachi with 100k simulations 90% of the time, and based on comparing to published stats, the net is much deeper and almost certainly much stronger than AlphaGoFan's (SL) policy net. It's probably deeper and stronger than Leela classic's policy net too. |
Author: | bernds [ Tue Apr 03, 2018 6:21 am ] |
Post subject: | Re: Where would you play? |
Author: | gowan [ Tue Apr 03, 2018 9:17 am ] |
Post subject: | Re: Where would you play? |
Reducing the center is clearly required. I would try to stay away from the Black super strong left side. Taking this into account I'd play on the 10 line maybe around M, N, or O. However, I'd not expect to win. That center is just too big and White's territory is too low. |
Author: | Knotwilg [ Tue Apr 03, 2018 9:25 am ] |
Post subject: | Re: Where would you play? |
Here we go: Edit: |
Author: | Bill Spight [ Tue Apr 03, 2018 10:28 am ] |
Post subject: | Re: Where would you play? |
What moves would I consider? OC, I immediately noticed the one space jumps. You always notice the one space jumps. ![]() OC, I also notice the peep against the Black group in the top right. You always notice peeps, too. But I doubt it would be answered by filling. ![]() |
Author: | jeromie [ Tue Apr 03, 2018 12:02 pm ] |
Post subject: | Re: Where would you play? |
I'm not a dan player, but |
Author: | drmwc [ Tue Apr 03, 2018 12:51 pm ] |
Post subject: | Re: Where would you play? |
Author: | Gomoto [ Tue Apr 03, 2018 2:47 pm ] |
Post subject: | Re: Where would you play? |
Author: | Uberdude [ Sat Apr 07, 2018 12:33 pm ] |
Post subject: | Re: Where would you play? |
Thanks everyone for the responses, interesting to see how you think. The move Carlo played was n10. A few people suggested moves here or in the vicinity so it doesn't seem particularly suspicious to me. Black chose to attach to the reduction rather than swallow it whole, so he converted a decent chunk of the right side to territory but also gave white momentum to extricate his stones, I think as white I would breathe a sigh of relief. To me black messed up the yose in not hassling the thinness of white around m6: that area should have been reduced to 0 points in sente but white ended up making 5+ there. Leela was also very critical of k10. Plus the r2 ko never happened. I thought Russian 6ds were generally excellent at yose, but doesn't seem the case here. |
Author: | Bill Spight [ Thu May 31, 2018 10:32 pm ] |
Post subject: | Re: Where would you play? |
I just had an interesting experience. I was playing over the game record without giving it much thought. To my surprise, when I got to the position that Uberdude had posed as a problem, the board looked different to me. Perhaps that had to do with my state of relaxation, but I also think that having seen White's move gave me a new perspective. Although I recognize that dealing with the opponent's thickness is a relative weakness of mine, because I am typically the one who makes thickness, I was not daunted by the opponent's thickness; instead, the right side and center looked quite big and inviting. ![]() This change in perspective was not the result of remembering exactly where White had played in the actual game (I didn't), nor of studying the subsequent play (I never did). Aside from my present state of relaxation, I think that simply seeing White's play altered my perception of the game, and that it will alter my perception of other positions, as well. Humans are quite adaptable, and one of our strengths, imitation, is not just a rote copycat ability. I think that modern go AI bots will have a great effect on the next generation of players, by revealing possibilities and widening their horizons. They will see the go board differently than our generation does. 20 years from now I would not be surprised to see a player emerge who is even better than Go Seigen was. |
Author: | Kirby [ Thu May 31, 2018 11:36 pm ] |
Post subject: | Re: Where would you play? |
My idea from the OP was to block the ko in the bottom right. I guess black probably wins from the moyo after that, though. |
Author: | Ian Butler [ Fri Jun 01, 2018 8:51 am ] |
Post subject: | Re: Where would you play? |
Bill Spight wrote: I just had an interesting experience. I was playing over the game record without giving it much thought. To my surprise, when I got to the position that Uberdude had posed as a problem, the board looked different to me. Perhaps that had to do with my state of relaxation, but I also think that having seen White's move gave me a new perspective. Although I recognize that dealing with the opponent's thickness is a relative weakness of mine, because I am typically the one who makes thickness, I was not daunted by the opponent's thickness; instead, the right side and center looked quite big and inviting. ![]() This change in perspective was not the result of remembering exactly where White had played in the actual game (I didn't), nor of studying the subsequent play (I never did). Aside from my present state of relaxation, I think that simply seeing White's play altered my perception of the game, and that it will alter my perception of other positions, as well. Humans are quite adaptable, and one of our strengths, imitation, is not just a rote copycat ability. I think that modern go AI bots will have a great effect on the next generation of players, by revealing possibilities and widening their horizons. They will see the go board differently than our generation does. 20 years from now I would not be surprised to see a player emerge who is even better than Go Seigen was. Interesting, I have the opposite view concerning moyos. I see them as territory too easily and really have to learn to see the many possibilites how to deal with them ![]() For me it's amazing white managed to win that game. Great game record, too. |
Author: | Kirby [ Fri Jun 01, 2018 9:17 am ] |
Post subject: | Re: Where would you play? |
I looked at this position with Lizzie using Leela Zero, and let it ponder for awhile. I don't have a GPU, so I don't know how meaningful it is, but the top move according to my computer was this: But then I played out the position for a few moves using the top suggestions each time from Leela Zero, and black killed white and the win percentage dropped... Maybe my computer doesn't have enough processing power? Seems weird to suggest a move where you'll die as the top move. Maybe I'm analyzing it wrong ![]() |
Author: | Bill Spight [ Fri Jun 01, 2018 9:44 am ] |
Post subject: | Re: Where would you play? |
Kirby wrote: Seems weird to suggest a move where you'll die as the top move. Maybe I'm analyzing it wrong ![]() How does LeelaZero evaluate the position after ![]() |
Page 1 of 2 | All times are UTC - 8 hours [ DST ] |
Powered by phpBB © 2000, 2002, 2005, 2007 phpBB Group http://www.phpbb.com/ |