A full game of Go with beginner-friendly commentary
Re: A full game of Go with beginner-friendly commentary
Black (1) runs towards open space while threatening to connect to his top right group. But how open is that space, really? Where are black's nearest friends? Now we see why our marked moves on the right were so useful. They not only reduced black's territory, they also helped our attack.
With (2) on the board, one more move in the top right might be enough to kill that group, but we aren't good enough to know for sure. If we fail to kill black's running group, we'll give that a shot.
Re: A full game of Go with beginner-friendly commentary
Scary stuff. Black found a good sequence in order to cut our top left group off from the marked stones. We had to protect our top left group because it, too, wasn't looking too healthy. Now it is, after effectively capturing (1), but with (3) black is isolating the marked stones. We should have played at A instead of at L14, to keep the stones more solidly connected. Black can cut them off from each other relatively easily now, but if he does this we can hopefully still connect to our top left group.
Re: A full game of Go with beginner-friendly commentary
Black indeed cut our stones off from each other with (3) and (5). Our (6) attempts to connect by going around black's stones, but black A is looking scary. If we lose our K16 group, it's probably game over. If we can save it, we're looking good. Now, we need to use our "reading" skills to find the best possible sequence. This is the part where the game is very chess-like, as it becomes all about looking ahead as far and as accurately as possible.
(of course, someone who is actually good at the game would've analysed this sequence before even playing L14)
Re: A full game of Go with beginner-friendly commentary
Black did play the move we expected, but we found a good answer in (2). We are connected now, and black has no way to connect his two weak groups. As predicted, our attack on black's G15 group has weakened the position of his M16 group, which now has nowhere to run. Our goal is unchanged: To kill either one of those groups. For now, black is still running. We can expect a move around A next.
Re: A full game of Go with beginner-friendly commentary
Some fighting moves. We end up being forced (I think) to connect our groups with (10), but we did take away all of black's possible ways to make a living group locally. He has to run again.
Re: A full game of Go with beginner-friendly commentary
But where is black running to? At this point, he's banking on creating enough weaknesses in the white position so that he can turn around and attack us. They do say offense is the best defense, and this rings quite true in Go as well.
We played (4) instead of cutting black off immediately, because if we had played at A he would end up connecting his two groups with this pesky sequence:
Re: A full game of Go with beginner-friendly commentary
Black played one more forcing move at (1), which we don't quite see the point of, then split our left side groups with (3). Perhaps his plan now is to isolate and attack our marked group in the middle by cutting at A at some point, but this seems completely unrealistic considering how much space it has to escape, and that it has friends near (the bottom left marked group).
I think we're probably winning at this point, unless I'm missing some crazy sequence where black escapes.
Re: A full game of Go with beginner-friendly commentary
Black (1) didn't seem very good, because it gave us time to fix our weakness with (4). Now, I really don't see a way out for him. He isn't any closer to making life locally, either.
Re: A full game of Go with beginner-friendly commentary
Game over
We win by resignation
Black made a final effort to escape but found no way out. He then resigned. Here's the final position with some markings to clarify things:
- Our territory is marked with X
- Black territory is marked with O
(this includes dead stones, which are worth 2 points).
(E16 is not a dead stone but the last played move).
There's still a decent amount of unclaimed space in the lower center and bottom center, but black judged he is too far behind and our position in that area is too strong for him to make up for the deficit. As we mentioned early on in the game, black's territorial style got him a lot of points along the edges but our strength in the center allowed us to strongly attack his group and eventually kill it off.
Now it's time to feed the game to the robot, and see where we messed up.
We win by resignation
Black made a final effort to escape but found no way out. He then resigned. Here's the final position with some markings to clarify things:
- Our territory is marked with X
- Black territory is marked with O
(this includes dead stones, which are worth 2 points).
(E16 is not a dead stone but the last played move).
There's still a decent amount of unclaimed space in the lower center and bottom center, but black judged he is too far behind and our position in that area is too strong for him to make up for the deficit. As we mentioned early on in the game, black's territorial style got him a lot of points along the edges but our strength in the center allowed us to strongly attack his group and eventually kill it off.
Now it's time to feed the game to the robot, and see where we messed up.
Re: A full game of Go with beginner-friendly commentary
On each move the AI shows, the top number is the win percentage of the player whose turn it is, and the bottom number is the number of positions the AI analyzed with that move on the board.
Interestingly, Go AIs analyze much less positions per second than Chess AI's do. This shows how much more important looking ahead is in Chess, and how much more important strategy is here. The AI's "policy network" chooses moves that might be good based on what it learned through playing against itself, and its "value network" analyzes the possible board states that follow. The policy network is much more important than it is in Chess, because without it the value network would have way too many moves to analyze. This is why Go AIs could never beat humans until machine learning (resulting in the policy network) was used by Deepmind.
The AI I use is Leela Zero, which is not AlphaGo but works the same way.
The white circle is where we played. According to the AI, our win chance at this point in the game is around 52% if we play the correct move (the blue move is the one the AI prefers). So it's still a very even game here. We played aggressively with the intention of attacking the black stone that was just played while pressuring black's corner. The AI move has the same idea, but does so without directly sacrificing the stone below.
There were plenty of moves from both players that the AI disagreed with in the opening, but they weren't as impactful and it evened out in the end. After our move here, our win chance became 42%. This was the first relatively big mistake.
Then there is this point, where black played the move we didn't expect, but should've expected. We thought something had to be done about his corner here, but were wrong. Extending was apparently much better, since our win chance went down to 32% after we played in the corner.
Here it's black's move, and we can see the AI considers his win chance to be about 70% if he plays the blue move. Black also underestimated this move though.
There are a few more moves played until we finally come back to play the blue move, at which point our win chance is back to 40%.
I'm glad to see the AI agrees about the circled move being important. Black left it unplayed, and went to invade us. This put us back at 48%.
... but our response was also a mistake, of course.
The AI doesn't like our response here, either. Interestingly, it looks like it wants us to reduce black's lower right territory now. I didn't even consider this.
After this exchange black is ahead again, and one move that keeps coming back is the blue one here. It looks like this was a great move that was not on either of our radars. Now that I look at it, it makes sense, but it's not trivial to see. I suppose players of our level can't be expected to see the urgency of such moves.
Here we finally play the move we'd been wanting to play for a while (see the white circle), but the AI doesn't really like it anymore. Classic.
Here, we should've attacked black's group while defending our own. This makes sense in hindsight, considering we ran into trouble with our top middle group later in the game.
Black has been making some mistakes and we are now solidly in the lead at 70%. As we mentioned before though, our move here wasn't correct. I'm surprised to see it's not all that bad though.
Here the game gets complicated so the AI can't really be relied on 100% anymore (we'd need to leave it running for a while to make sure it's not missing things), but it looks like black could've fixed all his problems with the blue move. He played 1 space to its right, which made all the difference.
From black 61% to white 70%. A 30% swing. The reasons I won't get into, suffice to say it's complicated.
In the fighting sequence that follows, almost every move is a mistake. That tends to happen when things get messy. Some interesting snapshots:
Here, black is somehow okay again, but the move he played (which I thought was reasonable) lowered his win chance by 20%.
Black missed the blue move here, which would've put him ahead. I do remember being scared of this move at the time.
This seemed to be black's final chance to escape and win the game. He had a move that would've put him at 75%, but what he played put us at 95%. That, we can safely say, was the game losing mistake.
In the moves that follow, the AI finds little to disagree with. Indeed, they were all fairly straightforward. Finally:
It's black's move, and the best move he has gives him a 2.2% win chance according to the AI. So I guess his best move was indeed to resign.
Interestingly, Go AIs analyze much less positions per second than Chess AI's do. This shows how much more important looking ahead is in Chess, and how much more important strategy is here. The AI's "policy network" chooses moves that might be good based on what it learned through playing against itself, and its "value network" analyzes the possible board states that follow. The policy network is much more important than it is in Chess, because without it the value network would have way too many moves to analyze. This is why Go AIs could never beat humans until machine learning (resulting in the policy network) was used by Deepmind.
The AI I use is Leela Zero, which is not AlphaGo but works the same way.
The white circle is where we played. According to the AI, our win chance at this point in the game is around 52% if we play the correct move (the blue move is the one the AI prefers). So it's still a very even game here. We played aggressively with the intention of attacking the black stone that was just played while pressuring black's corner. The AI move has the same idea, but does so without directly sacrificing the stone below.
There were plenty of moves from both players that the AI disagreed with in the opening, but they weren't as impactful and it evened out in the end. After our move here, our win chance became 42%. This was the first relatively big mistake.
Then there is this point, where black played the move we didn't expect, but should've expected. We thought something had to be done about his corner here, but were wrong. Extending was apparently much better, since our win chance went down to 32% after we played in the corner.
Here it's black's move, and we can see the AI considers his win chance to be about 70% if he plays the blue move. Black also underestimated this move though.
There are a few more moves played until we finally come back to play the blue move, at which point our win chance is back to 40%.
I'm glad to see the AI agrees about the circled move being important. Black left it unplayed, and went to invade us. This put us back at 48%.
... but our response was also a mistake, of course.
The AI doesn't like our response here, either. Interestingly, it looks like it wants us to reduce black's lower right territory now. I didn't even consider this.
After this exchange black is ahead again, and one move that keeps coming back is the blue one here. It looks like this was a great move that was not on either of our radars. Now that I look at it, it makes sense, but it's not trivial to see. I suppose players of our level can't be expected to see the urgency of such moves.
Here we finally play the move we'd been wanting to play for a while (see the white circle), but the AI doesn't really like it anymore. Classic.
Here, we should've attacked black's group while defending our own. This makes sense in hindsight, considering we ran into trouble with our top middle group later in the game.
Black has been making some mistakes and we are now solidly in the lead at 70%. As we mentioned before though, our move here wasn't correct. I'm surprised to see it's not all that bad though.
Here the game gets complicated so the AI can't really be relied on 100% anymore (we'd need to leave it running for a while to make sure it's not missing things), but it looks like black could've fixed all his problems with the blue move. He played 1 space to its right, which made all the difference.
From black 61% to white 70%. A 30% swing. The reasons I won't get into, suffice to say it's complicated.
In the fighting sequence that follows, almost every move is a mistake. That tends to happen when things get messy. Some interesting snapshots:
Here, black is somehow okay again, but the move he played (which I thought was reasonable) lowered his win chance by 20%.
Black missed the blue move here, which would've put him ahead. I do remember being scared of this move at the time.
This seemed to be black's final chance to escape and win the game. He had a move that would've put him at 75%, but what he played put us at 95%. That, we can safely say, was the game losing mistake.
In the moves that follow, the AI finds little to disagree with. Indeed, they were all fairly straightforward. Finally:
It's black's move, and the best move he has gives him a 2.2% win chance according to the AI. So I guess his best move was indeed to resign.
Re: A full game of Go with beginner-friendly commentary
-- deleted post --
Reason: on request (off-topic bulk delete)
Re: A full game of Go with beginner-friendly commentary
This game ended by resignation so you can't really count yet. There are no clear borders, after all. My markings were estimations.
When the game is done and there are clear borders, you count every empty space enclosed by your stones as 1 point, and every dead enemy stone encircled by yours as 2 points.
When the game is done and there are clear borders, you count every empty space enclosed by your stones as 1 point, and every dead enemy stone encircled by yours as 2 points.
Re: A full game of Go with beginner-friendly commentary
-- deleted post --
Reason: on request (off-topic bulk delete)
Re: A full game of Go with beginner-friendly commentary
I still haven't mastered checkers how am I supposed to learn THIS
Re: A full game of Go with beginner-friendly commentary
Throw checkers out the window and start learning this, that's how
Re: A full game of Go with beginner-friendly commentary
-- deleted post --
Reason: on request (off-topic bulk delete)
Re: A full game of Go with beginner-friendly commentary
Nice. Winning your first game is pretty unheard of lol.
It says I'm not allowed to access the review :( What's your username? Then I can look it up.
It says I'm not allowed to access the review :( What's your username? Then I can look it up.
Re: A full game of Go with beginner-friendly commentary
-- deleted post --
Reason: on request (off-topic bulk delete)
Re: A full game of Go with beginner-friendly commentary
Oh I'm still not able to access it. Must be some setting that made it private
You can probably switch the review to not private somehow
You can probably switch the review to not private somehow
-
- Jaeger
- Posts: 3107
- Joined: May 16, 2015
- ESO: Hyperactive Jam
Re: A full game of Go with beginner-friendly commentary
Winning against the 25k bot pretty easily now.
Re: A full game of Go with beginner-friendly commentary
Maybe incog will play you. He beat a 22k human
@kami_ryu
@kami_ryu
-
- Jaeger
- Posts: 3107
- Joined: May 16, 2015
- ESO: Hyperactive Jam
Re: A full game of Go with beginner-friendly commentary
Well 22 kilos isn't that much.
Re: A full game of Go with beginner-friendly commentary
Still not bad for a first game
- Mr_Bramboy
- Retired Contributor
- Posts: 8219
- Joined: Feb 26, 2015
- ESO: [VOC] Bram
- Location: Amsterdam
Re: A full game of Go with beginner-friendly commentary
I'm willing to put money on the fact that there are plenty of 22-kilo humans who would be able to beat me at Go.Jam wrote:Well 22 kilos isn't that much.
Re: A full game of Go with beginner-friendly commentary
-- deleted post --
Reason: on request (off-topic bulk delete)
Who is online
Users browsing this forum: No registered users and 11 guests