This is really one of the most memorable games of TwixtT I have ever played. Here are some comments (partly copied from the in-game chat). Post-game analysis was done with twixtbot-ui (https://github.com/stevens68/twixtbot-ui), where I made some contributions to recently. Please do check it out. It's an amazing tool, created by stevens68.
A game in two parts --- Looking back to this game, there are really two distinct parts: the first part, where shyryan was in the lead (up to 14.j18) and a second part, where I was in the lead (from 15.n15 onwards).
Up to 14.j18, things went quite smoothly for shyryan. AlphaZero gave him a 83% change to win at that point. All pegs at the bottom-right quarter of the bord are good positions, which don't leave much room to cross. There are enough opportunities to break through my two pegs on the H-column. After my move 7.r14, I commented: "Hmmm. This is not going in the right direction for me.... :-(".
In the beginning of the second part, I think I started with two or three pivotal moves from my side. First there is the 13.H16, which I really played with the mindset that I had "lost" on the right side of the board and should try the left (lower) side. Second, there are 15.N15 and 17.L16. After I more or less gave up on the left, I saw that 15.n15 would prevent shyryan to connect K16 and O14 directly, creating a possibility to break through.
My in-game comment after 15.n15 was: "I think I will try to delay your attack on the left side by first securing the connection between n15 and h16. Once I do that, I can go for the bottom on the left side (from h16) and reach for the top from n11..."
After 16.O18, shyryan replied: "That’s a good move! That plan makes sense, too. Good of you to cut me off in the middle and force me to really commit to the bottom. I agree that it looks like this will come down to the lower left. If you win there, I will really have a hard time stopping you on top. But if I win, you’ll have a hard time stopping me on the right." In hindsight, and with the analysis tools of AlphaZero, 16.N17 might have been a better move (see "worst move" section below).
Then, after 17.L16, I wrote: "This is really exciting! The last 2 moves and the next 2 will really decide the game, I think. This is the point where I think I really force you to choose: close that gap at h11/h16 or at j18/o18. The most exciting part is that the gap at j18/o18 is more strategical for you to close, but also more difficult to close (e.g.: takes more than 1 move)."
From that point, it was an exiting road to win, targeting first a link to the bottom via the left side (H-column) and then back to the M-column and back and forth. From move 17.L16 onward, AlphaZero calculates my change to win at around 80%.
Random other notes for some moves: --- - If being attacked, it sometimes helps to "take a step back" instead of making a forced decision to choose "left or right". A nice example of this is 11.O13, where I try to break through the N14 peg, splitting the two pegs on the K-column and the one at Q10. Instead of choosing a connection from O14 to either Q10 or one of the K-pegs, he plays: 12.P16. My comment on that was: "Nice move. That’s one of the “patterns” that often work: “take a step back” when attacked in the front. I’ll take a stab at the left now... See where that takes me...", after which I played 13.H16, which is in itself not a remarkable move, but it proofed to be an essential peg later on in the game.
- 24.F20 was a surprising move for me. Being in "a winning mood" I was totally surprised by that move and had to think hard to find an appropriate answer. My comments with my final response 25.F19 were: "This is a perfect example of why I love Twixt. At a glance, my first response is: oh no! he’s got me nailed! But when you look carefully, there is a specific path that leads you out of it. It is so extremely satisfying to find that path... Let’s see if I can hold my H11 position. In my evaluation, that’s the weakest point for me right now. The (0,5)-gap there feels big and if I were you, I’d think I would have an opportunity there attacking it by placing J14 and in the next move I11?"
- A similar surprise was presented by 28.F16. Shyryan's comments were: "I would like to try this move, actually! There’s a general principle of keeping your gaps separated from one another when possible; I was expecting G14 from you instead of G13 for that reason. It might not end up making a difference in this game, but it has given me a faint glimmer of hope." Also here, I found a satisfying response by playing 29.F15, creating four possible connections between my two segments on the F/G/H columns. I'd like to add here that placing connections parallel to each other (like F15/I15 and F19/I20) creates a very strong position, which is hard to stop. So: never be afraid to move perpendicular to the line between your two borders... But... After the brilliant 30.E18, I was again surprised by the double/triple threat. But also here, finally I found in 31.I22 a good response.
Relative player strength --- Analyzing the game with twistbot, I found that almost all of our moves correspond with moves that are also considered by the AlphaZero bot. The moves that we make are almost all in the top-3 moves that AlphaZero would have chosen as well. What's remarkable is that both statistics (percentage of moves in link with the bot's policy and the relative rank of the moves that we actually played) are exactly the same for both of us.
Worst moves --- As mentioned above, me breaking through the L/M/N column to the bottom at move 15 and 17 were pivotal. Evaluating the game, shyryan's move 16.O18 was rated as the move where he lost the most of his chance to win. Instead of that move, |16.N17 would have been better. A possible continuation would be: |16.N1717.L1618.L18. Not sure how that would play out, but it's certainly a better outcome for black.
As for my game, AlphaZero thinks that after 2.swap, I should have played on the E-line. Specifically E7/E12 are recommended moves. This will probably lead to some close-quarter battling on the top-left side of the board. Playing this line doesn't;t really mean that an actual connection is made to the top over the E-line, but it merely tries to place some strategically placed white pegs around the top-left area, isolating the G7 peg.
Final words --- I'd like to thank shyryan for an epic game and very insightful and open comments during the game.
Wow, yeah. I had written off the top side long ago because for a long time, whichever side I attacked, white could always just use the other side. I didn't realize that with my play at the bottom, it had kind of become "dealer's choice" down there, so whichever side white took on the top, I could address on the bottom. Neat!
Also thank you agtoever for an awesome game and for running it through the bot analysis! Wish I could have been the one to point out this fresh attack on the top side during the game. Next time. : )
On 2021-03-28 at 21:09, Alan Hensel said:
Yes, |32.k10 forces White to choose a line, but the line on the right looks more connectable to the top, so I'd choose it. For example, 33.m22. Then Black has to stop n11 from reaching the top.
If the bot switches sides after J18, that suggests J18 was a losing move. What does the bot think of |14.i11 instead?
On 2021-03-28 at 22:50, Peyrol said:
Sorry I didn't mean to be brusque. I would be grateful for any analysis, agtoever.
On 2021-04-28 at 22:09, agtoever said:
@Peyrol: thanks for your comments and suggestions.
- Indeed, TwixtBot finds 14.i11 a better move than 14.J18., but not with great margin. I think it's the next 2+2 moves were more pivotal than move 14 itself... - As for the game after move 31: I didn't really look into that. Not in my analysis afterwards, but also not during the game. Too much attention to the bottom half of the board... :-). If I look at your suggestions, 32.k10 is indeed a very strong move. Your suggested continuation: 33.j1034.l835.m736.h1737.k2138.n7 would indeed be something that I should have played.
Bottom line: white loses a turn for ensuring a connection to the bottom. This gives black a potentially winning upper hand at the top half of the board. I'm quite sure that if we had continued playing, shyryan would have won the game...
A game in two parts
---
Looking back to this game, there are really two distinct parts: the first part, where shyryan was in the lead (up to 14.j18) and a second part, where I was in the lead (from 15.n15 onwards).
Up to 14.j18, things went quite smoothly for shyryan. AlphaZero gave him a 83% change to win at that point. All pegs at the bottom-right quarter of the bord are good positions, which don't leave much room to cross. There are enough opportunities to break through my two pegs on the H-column. After my move 7.r14, I commented: "Hmmm. This is not going in the right direction for me.... :-(".
In the beginning of the second part, I think I started with two or three pivotal moves from my side. First there is the 13.H16, which I really played with the mindset that I had "lost" on the right side of the board and should try the left (lower) side. Second, there are 15.N15 and 17.L16. After I more or less gave up on the left, I saw that 15.n15 would prevent shyryan to connect K16 and O14 directly, creating a possibility to break through.
My in-game comment after 15.n15 was: "I think I will try to delay your attack on the left side by first securing the connection between n15 and h16. Once I do that, I can go for the bottom on the left side (from h16) and reach for the top from n11..."
After 16.O18, shyryan replied: "That’s a good move! That plan makes sense, too. Good of you to cut me off in the middle and force me to really commit to the bottom. I agree that it looks like this will come down to the lower left. If you win there, I will really have a hard time stopping you on top. But if I win, you’ll have a hard time stopping me on the right." In hindsight, and with the analysis tools of AlphaZero, 16.N17 might have been a better move (see "worst move" section below).
Then, after 17.L16, I wrote: "This is really exciting! The last 2 moves and the next 2 will really decide the game, I think. This is the point where I think I really force you to choose: close that gap at h11/h16 or at j18/o18. The most exciting part is that the gap at j18/o18 is more strategical for you to close, but also more difficult to close (e.g.: takes more than 1 move)."
From that point, it was an exiting road to win, targeting first a link to the bottom via the left side (H-column) and then back to the M-column and back and forth. From move 17.L16 onward, AlphaZero calculates my change to win at around 80%.
Random other notes for some moves:
---
- If being attacked, it sometimes helps to "take a step back" instead of making a forced decision to choose "left or right". A nice example of this is 11.O13, where I try to break through the N14 peg, splitting the two pegs on the K-column and the one at Q10. Instead of choosing a connection from O14 to either Q10 or one of the K-pegs, he plays: 12.P16. My comment on that was: "Nice move. That’s one of the “patterns” that often work: “take a step back” when attacked in the front. I’ll take a stab at the left now... See where that takes me...", after which I played 13.H16, which is in itself not a remarkable move, but it proofed to be an essential peg later on in the game.
- 24.F20 was a surprising move for me. Being in "a winning mood" I was totally surprised by that move and had to think hard to find an appropriate answer. My comments with my final response 25.F19 were: "This is a perfect example of why I love Twixt. At a glance, my first response is: oh no! he’s got me nailed! But when you look carefully, there is a specific path that leads you out of it. It is so extremely satisfying to find that path... Let’s see if I can hold my H11 position. In my evaluation, that’s the weakest point for me right now. The (0,5)-gap there feels big and if I were you, I’d think I would have an opportunity there attacking it by placing J14 and in the next move I11?"
- A similar surprise was presented by 28.F16. Shyryan's comments were: "I would like to try this move, actually! There’s a general principle of keeping your gaps separated from one another when possible; I was expecting G14 from you instead of G13 for that reason. It might not end up making a difference in this game, but it has given me a faint glimmer of hope." Also here, I found a satisfying response by playing 29.F15, creating four possible connections between my two segments on the F/G/H columns. I'd like to add here that placing connections parallel to each other (like F15/I15 and F19/I20) creates a very strong position, which is hard to stop. So: never be afraid to move perpendicular to the line between your two borders... But... After the brilliant 30.E18, I was again surprised by the double/triple threat. But also here, finally I found in 31.I22 a good response.
Relative player strength
---
Analyzing the game with twistbot, I found that almost all of our moves correspond with moves that are also considered by the AlphaZero bot. The moves that we make are almost all in the top-3 moves that AlphaZero would have chosen as well. What's remarkable is that both statistics (percentage of moves in link with the bot's policy and the relative rank of the moves that we actually played) are exactly the same for both of us.
Worst moves
---
As mentioned above, me breaking through the L/M/N column to the bottom at move 15 and 17 were pivotal. Evaluating the game, shyryan's move 16.O18 was rated as the move where he lost the most of his chance to win. Instead of that move, |16.N17 would have been better. A possible continuation would be: |16.N17 17.L16 18.L18. Not sure how that would play out, but it's certainly a better outcome for black.
As for my game, AlphaZero thinks that after 2.swap, I should have played on the E-line. Specifically E7/E12 are recommended moves. This will probably lead to some close-quarter battling on the top-left side of the board. Playing this line doesn't;t really mean that an actual connection is made to the top over the E-line, but it merely tries to place some strategically placed white pegs around the top-left area, isolating the G7 peg.
Final words
---
I'd like to thank shyryan for an epic game and very insightful and open comments during the game.
It looks a lot better than resigning.
Also thank you agtoever for an awesome game and for running it through the bot analysis! Wish I could have been the one to point out this fresh attack on the top side during the game. Next time. : )
34.l8 35.n7 36.m7 37.p8
34.m9 35.o9 36.o5 37.n7
34.n7 35.m9 36.k4 37.h5
Maybe there's a way, but I'm not seeing it.
- Indeed, TwixtBot finds 14.i11 a better move than 14.J18., but not with great margin. I think it's the next 2+2 moves were more pivotal than move 14 itself...
- As for the game after move 31: I didn't really look into that. Not in my analysis afterwards, but also not during the game. Too much attention to the bottom half of the board... :-). If I look at your suggestions, 32.k10 is indeed a very strong move. Your suggested continuation: 33.j10 34.l8 35.m7 36.h17 37.k21 38.n7 would indeed be something that I should have played.
Bottom line: white loses a turn for ensuring a connection to the bottom. This gives black a potentially winning upper hand at the top half of the board. I'm quite sure that if we had continued playing, shyryan would have won the game...