in the last diagram is always what happens if Black atari from below and all ladders are favourable for Black.Opening study with KataGo
-
Harleqin
- Lives in sente
- Posts: 921
- Joined: Sat Mar 06, 2010 10:31 am
- Rank: German 2 dan
- GD Posts: 0
- Has thanked: 401 times
- Been thanked: 164 times
Re: Opening study with KataGo
I think the interesting part about a wedge like in the last diagram is always what happens if Black atari from below and all ladders are favourable for Black.
in the last diagram is always what happens if Black atari from below and all ladders are favourable for Black.A good system naturally covers all corner cases without further effort.
-
hakuseki
- Dies with sente
- Posts: 100
- Joined: Thu Oct 29, 2020 6:18 am
- Rank: KGS 2 dan
- GD Posts: 0
- KGS: hakuseki
- Been thanked: 15 times
Game 28
I've taken a small break from studying for the last few weeks. Now that I'm back, I have a couple of new study ideas.
For a while I've been aiming for a rather peaceful style. However, I'd like to try playing more fighting games now. I'm hoping this will expose more mistakes that I can learn from. For example, in a very peaceful game my opponent and I might only play about 60 points' worth of mistakes each. Even so I am still only about KGS 2 dan. I imagine that to reach KGS 5 dan I should be able to correct about 39 points (i.e. three handicap stones) of those mistakes. But out of 60 points of mistakes, it is sometimes quite difficult to find 39 points of mistakes that I can effectively study and fix. In a fighting game, at my level, it's not strange to see 150 points or more of mistakes from each side, and many more of these mistakes seem fixable.
My second thought is that I'd like to try reviewing some games from stronger players, e.g. pro or AI games, in addition to my own games.
Here's one such game. KataGo has a (new, I think?) "playout doubling advantage" configuration option which causes it to assume it is stronger than its opponent and play more tricky moves. This is a game between two KataGo instances that both assume their opponent is weaker. Also, black thinks the komi is 13 and white thinks it is 0. So they were both quite aggressive.
Game 28, position 1 Nothing remarkable yet; a pretty normal beginning. I only know how to place numbered stones from
to 
in a diagram, btw. Is there a way to go further in a single diagram?
Game 28, position 2 If one focuses narrowly on the joseki, one might play b instead of
here, but is a better move as it is simultaneously responding to the approach on the upper left stone.
is a joseki choice I see fairly often from KataGo, but I still haven't used it much myself. However, from now I intend to try it. A couple quick variations:
Since is on the fourth line, it doesn't securely build territory. So 
is the sort of thing that would make me nervous. However, 
is a decent continuation that does build territory.
The kosumi-tsuke and jump is, however, about 0.3 points better.
Game 28, position 4 The double approach at a is to me a more obvious move than, and even receives a slightly better evaluation, but is fine and there are indeed several good moves in this position.
Now that white's three stones in the upper-right have been pincered, they will run towards the center, most likely through any of c through f. If black's marked stone were on the third line, white's strong preference would be for d, but in this case these four options are all considered good.
Game 28, position 5 It looks a bit like white is establishing a framework on the top, which could be continued with a capping move such as c, d, or e. But black is also building up the right side, so white might also like to invade with a move like a or b.
Game 29, position 6 White chooses to split black's moyo on the right rather than capping the black stone on the top side. Afterwards, black runs towards the center with. This shape surprised me, as I would normally only think of something like a. Actually both moves are good. seems a bit less effective when considered purely in terms of speed towards the center, but it simultaneously threatens a splitting move at b.
Game 29, position 7 Ignoring black's threat, white launches an attack with
. I think I would not have even recognized this as a weak point at this point in the game.
Black's defense at seems to be making miai of a and b. This is another move I probably would not have considered in favor of playing a or b directly. However, has the virtue of making a connection with white's right-side stone more difficult.
I was curious about what happens if black blocks on top, so I tried reading out this variation. Now white will either be able to connect on the right side or capture two stones in the corner.
I checked with KataGo, and is a slight mistake compared to a -- but still good enough to refute .
Game 29, position 8 White connects with this forcing sequence. The upper right is now a tangle of weak groups and black has the opportunity to do something.
Game 29, position 9 First, black plays two forcing moves to help solidify the corner in sente. It would be quite reasonable to play one more move at a, but black chooses to play that a bit later instead.
Game 29, position 10 Next, black makes shape with and before forcing white's connection with 
.
In review, KataGo thinks was a slight mistake -- the 
exchange should have been played first. Let's try to see why.
First let's consider what would happen if black cuts instead of extending with. White will be able to play this sequeezing sequence, finally connecting with a and b and making a tiger's mouth connection with c. This is slightly better for white than the game result.
If black has exchanged first, however, then white is short a liberty. Now if white tries to deal with the cut in the same way, the result would be a complete disaster.
This is a better way of dealing with the cut. White will live small while black gets influence. However, black is still winning the game by over 9 points if this happens. Actually, being cut by is unacceptable no matter how white plays.
Instead, white should play the one-space jump at, resulting in this sequence where white finishes in gote locally with 
. Black has gained about 0.7 points over the actual game result.
Now, if we scroll all the way back to position 10, I think we can see why should have been at -- it makes white's hane at nonviable and forces white to take gote.
Game 28, position 11 (preview) Not sure if I'll continue the review, but here's a preview of the next few moves in case anyone's curious. I have a KGS game that I'd like to review next, so I might just move on to that.
For a while I've been aiming for a rather peaceful style. However, I'd like to try playing more fighting games now. I'm hoping this will expose more mistakes that I can learn from. For example, in a very peaceful game my opponent and I might only play about 60 points' worth of mistakes each. Even so I am still only about KGS 2 dan. I imagine that to reach KGS 5 dan I should be able to correct about 39 points (i.e. three handicap stones) of those mistakes. But out of 60 points of mistakes, it is sometimes quite difficult to find 39 points of mistakes that I can effectively study and fix. In a fighting game, at my level, it's not strange to see 150 points or more of mistakes from each side, and many more of these mistakes seem fixable.
My second thought is that I'd like to try reviewing some games from stronger players, e.g. pro or AI games, in addition to my own games.
Here's one such game. KataGo has a (new, I think?) "playout doubling advantage" configuration option which causes it to assume it is stronger than its opponent and play more tricky moves. This is a game between two KataGo instances that both assume their opponent is weaker. Also, black thinks the komi is 13 and white thinks it is 0. So they were both quite aggressive.
Game 28, position 1 Nothing remarkable yet; a pretty normal beginning. I only know how to place numbered stones from
to in a diagram, btw. Is there a way to go further in a single diagram?Game 28, position 2 If one focuses narrowly on the joseki, one might play b instead of
here, but is a better move as it is simultaneously responding to the approach on the upper left stone. is a joseki choice I see fairly often from KataGo, but I still haven't used it much myself. However, from now I intend to try it. A couple quick variations:Since
is on the fourth line, it doesn't securely build territory. So is the sort of thing that would make me nervous. However, is a decent continuation that does build territory.The kosumi-tsuke and jump is, however, about 0.3 points better.
Game 28, position 4 The double approach at a is to me a more obvious move than
, and even receives a slightly better evaluation, but is fine and there are indeed several good moves in this position.Now that white's three stones in the upper-right have been pincered, they will run towards the center, most likely through any of c through f. If black's marked stone were on the third line, white's strong preference would be for d, but in this case these four options are all considered good.
Game 28, position 5 It looks a bit like white is establishing a framework on the top, which could be continued with a capping move such as c, d, or e. But black is also building up the right side, so white might also like to invade with a move like a or b.
Game 29, position 6 White chooses to split black's moyo on the right rather than capping the black stone on the top side. Afterwards, black runs towards the center with
. This shape surprised me, as I would normally only think of something like a. Actually both moves are good. seems a bit less effective when considered purely in terms of speed towards the center, but it simultaneously threatens a splitting move at b.Game 29, position 7 Ignoring black's threat, white launches an attack with
. I think I would not have even recognized this as a weak point at this point in the game.Black's defense at
seems to be making miai of a and b. This is another move I probably would not have considered in favor of playing a or b directly. However, has the virtue of making a connection with white's right-side stone more difficult.I was curious about what happens if black blocks on top, so I tried reading out this variation. Now white will either be able to connect on the right side or capture two stones in the corner.
I checked with KataGo, and
is a slight mistake compared to a -- but still good enough to refute .Game 29, position 8 White connects with this forcing sequence. The upper right is now a tangle of weak groups and black has the opportunity to do something.
Game 29, position 9 First, black plays two forcing moves to help solidify the corner in sente. It would be quite reasonable to play one more move at a, but black chooses to play that a bit later instead.
Game 29, position 10 Next, black makes shape with
and before forcing white's connection with .In review, KataGo thinks
was a slight mistake -- the exchange should have been played first. Let's try to see why.First let's consider what would happen if black cuts instead of extending with
. White will be able to play this sequeezing sequence, finally connecting with a and b and making a tiger's mouth connection with c. This is slightly better for white than the game result.If black has exchanged
first, however, then white is short a liberty. Now if white tries to deal with the cut in the same way, the result would be a complete disaster.This is a better way of dealing with the cut. White will live small while black gets influence. However, black is still winning the game by over 9 points if this happens. Actually, being cut by
is unacceptable no matter how white plays.Instead, white should play the one-space jump at
, resulting in this sequence where white finishes in gote locally with . Black has gained about 0.7 points over the actual game result.Now, if we scroll all the way back to position 10, I think we can see why
should have been at -- it makes white's hane at nonviable and forces white to take gote.Game 28, position 11 (preview) Not sure if I'll continue the review, but here's a preview of the next few moves in case anyone's curious. I have a KGS game that I'd like to review next, so I might just move on to that.
-
hakuseki
- Dies with sente
- Posts: 100
- Joined: Thu Oct 29, 2020 6:18 am
- Rank: KGS 2 dan
- GD Posts: 0
- KGS: hakuseki
- Been thanked: 15 times
Game 29
This is a game I played online. I played black in this game. My mistakes totaled about 13 points by move 80 and 185 points by the game end (331 moves). Most of the mistakes were during the life and death struggles in the midgame.
Game 29, position 1 Actually, I think black a is a pretty obvious intuition. Yet somehow it felt too submissive so I played b instead, which was a mistake.
Perhaps I was worried about how black should continue after. Something like b, c, or d feels quite slow, and if black plays any of these moves then the exchange would be good for white. Actually, black should extend at a.
Game 29, position 2 I didn't want to have to answer an atari, so I extended at b. This was a mistake!
If black cuts, the atari of would not be sente. Black will just capture two stones with .
Game 29, position 3 The best move here is a. Black b would be losing about half a point, and c would lose 2.8 points. Unfortunately I played c. Although a would be the best move, I'll analyze b vs. c since these moves are easier for me to compare.
Here's what happened in the game -- bad result for black.
If black plays the hane at instead, this is one possible result. This is obviously a small improvement over the previous diagram.
If black has anything to fear about playing, it is this sequence. However, is gote. Although 
at a is a good move, this is also a perfectly reasonable time for black to tenuki.
Game 29, position 4 There are mistakes on both sides in this sequence, but black is worse. I think I saw as trying to connect to white's right side stone, and played to prevent the connection. But such a plan is almost useless. Black's two groups are both strong, so there is no need for black to defend here, and surrounding one stone is obviously small. Meanwhile white is getting great influence.
Instead of, there are big moves available like a, b, or c.
Game 29, position 1 Actually, I think black a is a pretty obvious intuition. Yet somehow it felt too submissive so I played b instead, which was a mistake.
Perhaps I was worried about how black should continue after
. Something like b, c, or d feels quite slow, and if black plays any of these moves then the exchange would be good for white. Actually, black should extend at a.Game 29, position 2 I didn't want to have to answer an atari, so I extended at b. This was a mistake!
If black cuts, the atari of
would not be sente. Black will just capture two stones with .Game 29, position 3 The best move here is a. Black b would be losing about half a point, and c would lose 2.8 points. Unfortunately I played c. Although a would be the best move, I'll analyze b vs. c since these moves are easier for me to compare.
Here's what happened in the game -- bad result for black.
If black plays the hane at
instead, this is one possible result. This is obviously a small improvement over the previous diagram.If black has anything to fear about playing
, it is this sequence. However, is gote. Although at a is a good move, this is also a perfectly reasonable time for black to tenuki.Game 29, position 4 There are mistakes on both sides in this sequence, but black is worse. I think I saw
as trying to connect to white's right side stone, and played to prevent the connection. But such a plan is almost useless. Black's two groups are both strong, so there is no need for black to defend here, and surrounding one stone is obviously small. Meanwhile white is getting great influence.Instead of
, there are big moves available like a, b, or c.-
hakuseki
- Dies with sente
- Posts: 100
- Joined: Thu Oct 29, 2020 6:18 am
- Rank: KGS 2 dan
- GD Posts: 0
- KGS: hakuseki
- Been thanked: 15 times
Shogi
This is a bit off-topic, but recently I've been studying shogi. A good number of people around me play shogi, and I'd like to be able to challenge them.
A couple months ago I stumbled across a video whose author described how he reached shodan. There were essentially three steps:
Tsumeshogi are kind of like the shogi equivalent of tsumego, but they're also a bit more specialized. There are tsumego for living and tsumego for killing, but tsumeshogi are always about killing. There's also a restriction that every move you play must be a checking move, presumably to avoid yielding sente which may allow the opponent to win the game elsewhere on the board.
One thing I am really enjoying is the lack of mistakes in the problems (actually, I have seen a couple typos, but it was very clear that these were simply typos and not conceptual errors). This is really different from my experience with go problems, where I've come to expect 3% or more of problems to contain errors in their solutions. I asked a friend about this, and he thinks that this is because tsumeshogi creators sign their problems. Indeed, the book I'm reading contains problems by many different authors, but the author's names are written above the problems. My friend suggests the authors would lose prestige if their problems were composed incorrectly.
I do feel a bit inspired to do more tsumego when I switch back to focusing on Go, but I'd like to put a bit of effort into choosing a problem collection without errors. I think in my game reviews I did find a sufficiency of situations where a player with moderately stronger reading skills should have been able to play the moves that I missed.
A couple months ago I stumbled across a video whose author described how he reached shodan. There were essentially three steps:
- Study tsumeshogi books
- Study a tsugi no itte book
- Play games with a stronger player
Tsumeshogi are kind of like the shogi equivalent of tsumego, but they're also a bit more specialized. There are tsumego for living and tsumego for killing, but tsumeshogi are always about killing. There's also a restriction that every move you play must be a checking move, presumably to avoid yielding sente which may allow the opponent to win the game elsewhere on the board.
One thing I am really enjoying is the lack of mistakes in the problems (actually, I have seen a couple typos, but it was very clear that these were simply typos and not conceptual errors). This is really different from my experience with go problems, where I've come to expect 3% or more of problems to contain errors in their solutions. I asked a friend about this, and he thinks that this is because tsumeshogi creators sign their problems. Indeed, the book I'm reading contains problems by many different authors, but the author's names are written above the problems. My friend suggests the authors would lose prestige if their problems were composed incorrectly.
I do feel a bit inspired to do more tsumego when I switch back to focusing on Go, but I'd like to put a bit of effort into choosing a problem collection without errors. I think in my game reviews I did find a sufficiency of situations where a player with moderately stronger reading skills should have been able to play the moves that I missed.
-
RobertJasiek
- Judan
- Posts: 6272
- Joined: Tue Apr 27, 2010 8:54 pm
- GD Posts: 0
- Been thanked: 797 times
- Contact:
Re: Shogi
What is "tsugi no itte", please?
See also my reply: https://www.lifein19x19.com/viewtopic.php?f=17&t=18266
See also my reply: https://www.lifein19x19.com/viewtopic.php?f=17&t=18266
-
hakuseki
- Dies with sente
- Posts: 100
- Joined: Thu Oct 29, 2020 6:18 am
- Rank: KGS 2 dan
- GD Posts: 0
- KGS: hakuseki
- Been thanked: 15 times
Re: Shogi
"Tsugi no itte" means "the next move." I haven't studied too deeply yet, but it seems like books in this genre have a great variety of whole-board problems.RobertJasiek wrote:What is "tsugi no itte", please?
See also my reply: https://www.lifein19x19.com/viewtopic.php?f=17&t=18266
-
hakuseki
- Dies with sente
- Posts: 100
- Joined: Thu Oct 29, 2020 6:18 am
- Rank: KGS 2 dan
- GD Posts: 0
- KGS: hakuseki
- Been thanked: 15 times
Game 30
Game 30
I'm going to try something different. I think my opening practice has improved my opening a lot. I'd like to see similar improvements to my midgame and endgame. However, playing a full game would be a bit tedious if I want to focus on e.g. endgame. So I would like to start my games from later positions.
Instead of starting from a different midgame or endgame position every time, however, I want to play repeatedly from the same starting position until I see major improvement. Then I will move on to a different starting position.
I've used KataGo to generate the following game. By the way, this is using stone scoring rules (i.e. with group tax), so the strategy is a bit different compared to modern Japanese or Chinese rules. If you're studying along with me, just beware as the lessons may not translate perfectly to games without group tax.
My study plan from here is as follows:
I'm going to try something different. I think my opening practice has improved my opening a lot. I'd like to see similar improvements to my midgame and endgame. However, playing a full game would be a bit tedious if I want to focus on e.g. endgame. So I would like to start my games from later positions.
Instead of starting from a different midgame or endgame position every time, however, I want to play repeatedly from the same starting position until I see major improvement. Then I will move on to a different starting position.
I've used KataGo to generate the following game. By the way, this is using stone scoring rules (i.e. with group tax), so the strategy is a bit different compared to modern Japanese or Chinese rules. If you're studying along with me, just beware as the lessons may not translate perfectly to games without group tax.
My study plan from here is as follows:
- Memorize this kifu (I'm up to about move 160 so far)
- Post an initial review of the game, including effects of group tax
- Split the game into sections of about 30–40 moves each (opening, early midgame, etc.)
- Play two mini-games (one as black, one as white) from the starting position of each section, against an AI, stopping after 30-40 moves
- Maybe post some review highlights from these games
- Calculate my total point loss, averaged over the black and white games
- Determine a goal (i.e. total point loss below some value x)
- Play mini-games, focusing on one game section at a time, until the goal is reached
-
hakuseki
- Dies with sente
- Posts: 100
- Joined: Thu Oct 29, 2020 6:18 am
- Rank: KGS 2 dan
- GD Posts: 0
- KGS: hakuseki
- Been thanked: 15 times
Game 30 comments
OK, here are some initial comments on game 30. I had a few unresolved questions; perhaps I will gain some insight into them in my follow-up studies. If anyone else would like to comment, I'd be equally happy to receive either helpful analysis or additional questions (which I may keep in the back of my mind as I continue to study this kifu).
-
hakuseki
- Dies with sente
- Posts: 100
- Joined: Thu Oct 29, 2020 6:18 am
- Rank: KGS 2 dan
- GD Posts: 0
- KGS: hakuseki
- Been thanked: 15 times
Game 30 stages
I'll split the game into the following stages:
The diagonal fuseki used in moves 1–4 was a prescribed opening in ancient Chinese Go. White also had the first move in those days, but I have not bothered to replicate that here. Anyway, I want to focus on this diagonal opening so I will start my practice games from move 5.
Code: Select all
initial moves: 1–4
opening: 5–46
late opening/early midgame: 47–80
midgame: 81–111
ōyose: 112–145
early endgame: 146–174
endgame: 175–210
late endgame and dame: 211–243
-
hakuseki
- Dies with sente
- Posts: 100
- Joined: Thu Oct 29, 2020 6:18 am
- Rank: KGS 2 dan
- GD Posts: 0
- KGS: hakuseki
- Been thanked: 15 times
Game 31
I played this opening up to move 45 as black. This game was played with group tax, so opening theory is slightly different than in modern Go without group tax. I lost 6.9 points in this opening.
Game 31, position 1 Analogously to answering a large knight's move with a large knight's move, I thought I'd try answering white's two-space jump with a two-space jump of my own. However, this does not seem to be a joseki. Any of a, b, c, or d is preferable; e is also not bad if I want to play a move close to the one I played.
Game 31, position 2 After, I wanted to help my touched stone, but I thought a would be too heavy. I tried . Actually, tenuki may be best, but the following local variation is also okay:
Game 31, position 3 Should black continue at a or b?
I did read that a seems to work. However, by this point in the game a ko fight was developing elsewhere, and I worried that a would leave white too many threats, so I played b. This was a mistake, as a is really a much stronger move.
Game 31, position 1 Analogously to answering a large knight's move with a large knight's move, I thought I'd try answering white's two-space jump with a two-space jump of my own. However, this does not seem to be a joseki. Any of a, b, c, or d is preferable; e is also not bad if I want to play a move close to the one I played.
Game 31, position 2 After
, I wanted to help my touched stone, but I thought a would be too heavy. I tried . Actually, tenuki may be best, but the following local variation is also okay:Game 31, position 3 Should black continue at a or b?
I did read that a seems to work. However, by this point in the game a ko fight was developing elsewhere, and I worried that a would leave white too many threats, so I played b. This was a mistake, as a is really a much stronger move.
-
hakuseki
- Dies with sente
- Posts: 100
- Joined: Thu Oct 29, 2020 6:18 am
- Rank: KGS 2 dan
- GD Posts: 0
- KGS: hakuseki
- Been thanked: 15 times
Game 32
I played this game from move 47 up to move 79 as black, and lost about 5.9 points. Actually, I played a bit further and made a large mistake on move 81; however, since my study plan calls for this playout to cover moves 47–80, I am not including that mistake in this review. I will add it to my problem collection instead.
Game 32, position 1 I expected white to push through with at , but instead white played this clamp. I thought this move looked like a mistake, but it's only losing about 0.3 points. I felt pretty unsure about my response at , but it is actually okay; a is slightly worse and reverts to the same sequence as if at .
Game 32, position 2 With, black's group is too thin; black should defend at a instead. By reinforcing in this way, black would be strong enough to play 
; but since black did not play a in the game, is too weak and black should have played at c instead.
is also a mistake. White should push through at b and fight.
Game 32, position 3 After, I considered cutting off white's stone by playing at a, but I thought the fight might be bad for me so I played instead. Black's bottom group is a bit weak after , which I think increases the value of b, so 
instead of b is also a mistake.
Variation Black can build some pretty nice territory by cutting with and instead.
Game 32, position 4 After , white tenukis with . This inconsistent play is not great for white, but what should black do?
Actually, capturing three stones with a would be a fine move for black, even though white will be able to live on the top.
My move was b, but this is losing 1.4 points and is worse than c or d. I was only thinking of ensuring dies while preserving territory on the top, but this would allow white to aim for the center:
Variation b By playing like this, white may profit from black's slack move at. In the game, white played at a instead of , which is a mistake. There will be a followup post about this in my problem collection thread.
Variation c With, black maintains access to the center. This is more important than capturing the marked stones. For example, a white move at a may simply be ignored. White moves such as b, c, or d, however, which threaten the connection of black's top groups or access to the center, would merit some response.
Variation d This is not the best move, but I include it because it is the best move with a similar intention to the move I played. The sequence shown here is not white's best sequence, but is simply showing how black might deal with some white overplays. If white cuts at , black maintains access to the center with and . White should now sacrifice the two marked stones, but if they stubbornly cling to life, black may kill white's corner instead with the sequence to b.
Game 32, position 1 I expected white to push through with
at , but instead white played this clamp. I thought this move looked like a mistake, but it's only losing about 0.3 points. I felt pretty unsure about my response at , but it is actually okay; a is slightly worse and reverts to the same sequence as if at .Game 32, position 2 With
, black's group is too thin; black should defend at a instead. By reinforcing in this way, black would be strong enough to play ; but since black did not play a in the game, is too weak and black should have played at c instead. is also a mistake. White should push through at b and fight.Game 32, position 3 After
, I considered cutting off white's stone by playing at a, but I thought the fight might be bad for me so I played instead. Black's bottom group is a bit weak after , which I think increases the value of b, so instead of b is also a mistake.Variation Black can build some pretty nice territory by cutting with
and instead.Game 32, position 4 After
, white tenukis with . This inconsistent play is not great for white, but what should black do?Actually, capturing three stones with a would be a fine move for black, even though white will be able to live on the top.
My move was b, but this is losing 1.4 points and is worse than c or d. I was only thinking of ensuring
dies while preserving territory on the top, but this would allow white to aim for the center:Variation b By playing like this, white may profit from black's slack move at
. In the game, white played at a instead of , which is a mistake. There will be a followup post about this in my problem collection thread.Variation c With
, black maintains access to the center. This is more important than capturing the marked stones. For example, a white move at a may simply be ignored. White moves such as b, c, or d, however, which threaten the connection of black's top groups or access to the center, would merit some response.Variation d This
is not the best move, but I include it because it is the best move with a similar intention to the move I played. The sequence shown here is not white's best sequence, but is simply showing how black might deal with some white overplays. If white cuts at , black maintains access to the center with and . White should now sacrifice the two marked stones, but if they stubbornly cling to life, black may kill white's corner instead with the sequence to b.-
hakuseki
- Dies with sente
- Posts: 100
- Joined: Thu Oct 29, 2020 6:18 am
- Rank: KGS 2 dan
- GD Posts: 0
- KGS: hakuseki
- Been thanked: 15 times
Game 33
I played this game from move 81 to move 111 as black. I lost about 16.8 points.
Game 33, position 1 My initial move is copied directly from KataGo's game. In that game, white responded at d, but this time my opponent played .
At this point, black might tenuki to play a big move elsewhere, like a or b. The local move c is good, but I thought this might be aji keshi, so I played d. I think this was pretty bad thinking. It's clear that black has worse shape problems than white, so playing c to solidify both sides benefits black more.
Here's the local continuation. Actually, many moves elsewhere were interposed in this sequence, but I omit those here and show the local sequence only, in which black and white did play alternating moves. I've mentioned was a mistake.
would be better placed at . Actually, under modern rules either move is viable. However, since I am playing with group tax, I should prefer the move that keeps my stones connected.
was played at , but this is living small in gote. It would have been better to play at a and fight the ko.
Game 33, position 2 Again I copied KataGo's move with. In the kifu I studied, white responded at a, but my opponent played 
instead.
Now black's shape has a defect. I chose to repair it with a solid connection at c. This is fine, but b may be very slightly better, and generally I think the shape looks much nicer, so I wish I had played b. Game 33, position 3 White's cutting stone is poised to cause great trouble for black. I basically ignored it to play b, perhaps thinking this was some kind of sente move, and I could come back and capture the cutting stone. My opponent even let me get away with this somewhat. But if white simply extends at a then black is cut to pieces. Black must prevent this by playing a immediately.
Game 33, position 1 My initial move
is copied directly from KataGo's game. In that game, white responded at d, but this time my opponent played .At this point, black might tenuki to play a big move elsewhere, like a or b. The local move c is good, but I thought this might be aji keshi, so I played d. I think this was pretty bad thinking. It's clear that black has worse shape problems than white, so playing c to solidify both sides benefits black more.
Here's the local continuation. Actually, many moves elsewhere were interposed in this sequence, but I omit those here and show the local sequence only, in which black and white did play alternating moves. I've mentioned
was a mistake. would be better placed at . Actually, under modern rules either move is viable. However, since I am playing with group tax, I should prefer the move that keeps my stones connected. was played at , but this is living small in gote. It would have been better to play at a and fight the ko.Game 33, position 2 Again I copied KataGo's move with
. In the kifu I studied, white responded at a, but my opponent played instead.Now black's shape has a defect. I chose to repair it with a solid connection at c. This is fine, but b may be very slightly better, and generally I think the shape looks much nicer, so I wish I had played b. Game 33, position 3 White's cutting stone is poised to cause great trouble for black. I basically ignored it to play b, perhaps thinking this was some kind of sente move, and I could come back and capture the cutting stone. My opponent even let me get away with this somewhat. But if white simply extends at a then black is cut to pieces. Black must prevent this by playing a immediately.
-
hakuseki
- Dies with sente
- Posts: 100
- Joined: Thu Oct 29, 2020 6:18 am
- Rank: KGS 2 dan
- GD Posts: 0
- KGS: hakuseki
- Been thanked: 15 times
Game 34
Game 34 notes
I played moves 112–145 as black and lost 11.7 points. This was the ōyose stage of the game. I'm not bothering with a detailed review of this playout for now, but I have a couple quick comments:
First, recall this sagari I played in game 33:
After this exchange, a descent at a is no longer a threat to white's corner. So black should omit. I didn't really notice this even during the review of game 33 and ended up repeating the mistake in this game.
Pseudo-miai values (coin values)
Aside from this, in game 34 I made several mistakes in judging the value of moves. This brought to mind an earlier project of mine called Miai values in the opening.
(As pointed out by both Mr. Jasiek and Mr. Spight, the values I elicited from KataGo do not fit the definition of miai values. I will instead call them "coin values" for the time being.)
I analyzed my mistakes from this game in terms of coin values. For each mistake I chose a nice-looking move suggested by KataGo and evaluated the coin values for my move and KataGo's move. In two cases my move had a higher coin value. These two moves account for 1.6 points lost. So I could have impoved by score by 10.1 points by playing KataGo's suggestions which had higher coin values.
I acknowledge this experiment could have been more scientific, as I only bothered checking the moves where I made a mistake, and only compared them against good moves chosen by KataGo. Still I'm quite convinced that coin values can be an effective heuristic for comparing gote moves in different regions of the board (which pretty much describes the ōyose problems in this particular game).
Therefore I am planning to continue investigating coin values. My current plan has two components:
Evaluate the usefulness of coin values. To do this I will put together a collection of test positions. These must be calm positions as I do not intend to use coin values as a heuristic when fighting. For each position, I will list the moves that seem plausible to me, and also choose a single move which I would play. Then I will compute coin values for each plausible move, and calculate the average point gain (or loss) of the moves with the highest coin value compared to the moves chosen by my intuition. This experiment will tell me whether coin values are a better heuristic than my current intuition for these kinds of positions.
Develop an intuition for coin values. I will do this by putting together a set of estimation problems. The task will be to guess the coin value of a given move in its whole board context. I believe I can control the problem difficulty somewhat by arranging the problems into sets with different value distributions. For example, an easy problem set might only include moves with values very close to either 4, 8, or 12 points (±0.1 points, for example). I would hope it's relatively easy to distinguish moves of small, medium, and large size like this. A more difficult problem set might contain values close to any even number, or any whole number, and so on.
I played moves 112–145 as black and lost 11.7 points. This was the ōyose stage of the game. I'm not bothering with a detailed review of this playout for now, but I have a couple quick comments:
First, recall this sagari I played in game 33:
After this exchange, a descent at a is no longer a threat to white's corner. So black should omit
. I didn't really notice this even during the review of game 33 and ended up repeating the mistake in this game.Pseudo-miai values (coin values)
Aside from this, in game 34 I made several mistakes in judging the value of moves. This brought to mind an earlier project of mine called Miai values in the opening.
(As pointed out by both Mr. Jasiek and Mr. Spight, the values I elicited from KataGo do not fit the definition of miai values. I will instead call them "coin values" for the time being.)
I analyzed my mistakes from this game in terms of coin values. For each mistake I chose a nice-looking move suggested by KataGo and evaluated the coin values for my move and KataGo's move. In two cases my move had a higher coin value. These two moves account for 1.6 points lost. So I could have impoved by score by 10.1 points by playing KataGo's suggestions which had higher coin values.
I acknowledge this experiment could have been more scientific, as I only bothered checking the moves where I made a mistake, and only compared them against good moves chosen by KataGo. Still I'm quite convinced that coin values can be an effective heuristic for comparing gote moves in different regions of the board (which pretty much describes the ōyose problems in this particular game).
Therefore I am planning to continue investigating coin values. My current plan has two components:
Evaluate the usefulness of coin values. To do this I will put together a collection of test positions. These must be calm positions as I do not intend to use coin values as a heuristic when fighting. For each position, I will list the moves that seem plausible to me, and also choose a single move which I would play. Then I will compute coin values for each plausible move, and calculate the average point gain (or loss) of the moves with the highest coin value compared to the moves chosen by my intuition. This experiment will tell me whether coin values are a better heuristic than my current intuition for these kinds of positions. Develop an intuition for coin values. I will do this by putting together a set of estimation problems. The task will be to guess the coin value of a given move in its whole board context. I believe I can control the problem difficulty somewhat by arranging the problems into sets with different value distributions. For example, an easy problem set might only include moves with values very close to either 4, 8, or 12 points (±0.1 points, for example). I would hope it's relatively easy to distinguish moves of small, medium, and large size like this. A more difficult problem set might contain values close to any even number, or any whole number, and so on.-
hakuseki
- Dies with sente
- Posts: 100
- Joined: Thu Oct 29, 2020 6:18 am
- Rank: KGS 2 dan
- GD Posts: 0
- KGS: hakuseki
- Been thanked: 15 times
Re: Opening study with KataGo
I've played through the endgame. I made some mistakes, but I don't think the detailed writeup is very interesting. Here are my overall statistics:
In the late endgame I actually gained points (in terms of KataGo's evaluation). Still I did make a mistake that I think warrants a remark, if not a diagram, simply because it is a novel concept to me: with an odd number of dame on the board, I chose to connect a simple ko instead of filling a dame. This would have been correct under Japanese rules, but, given that I had a surplus of ko threats, it turned out to be a mistake with stone scoring (which is similar to area scoring in this regard).
Throughout the entire game I lost 58.1 points. Next, I must play through the game as white before deciding my improvement goals. Still I have two ideas for improvement goals:
Reduce my mistakes by 28 points. This is roughly equivalent to two handicap stones, which is also suggestive of gaining two stones of strength. I'm not sure how well this skill would generalize beyond the special starting positions I'm using, or from AI to human opponents, but even 25% of two stones in strength would be very significant.
Reduce my mistakes by a factor of two. In this case that would mean reducing my mistakes by about 29 points, which is nearly the same as . Abstractly, a factor of 2 might make more sense than a fixed point-value improvement like 28, as some games are much messier than others, and I should expect that more improvement is possible in such messy games.
Anyway, I have already played the next opening as white, so I will move on to writing up some joseki mistakes next.
Code: Select all
Stage Moves Loss
initial moves 1–4 N/A
opening 5–46 6.9
late opening/early midgame 47–80 5.9
midgame 81–111 16.8
ōyose 112–145 11.7
early endgame 146–174 11.1
endgame 175–210 8.5
late endgame and dame 211–243 -2.8
TOTAL 1-243 58.1
Throughout the entire game I lost 58.1 points. Next, I must play through the game as white before deciding my improvement goals. Still I have two ideas for improvement goals:
Reduce my mistakes by 28 points. This is roughly equivalent to two handicap stones, which is also suggestive of gaining two stones of strength. I'm not sure how well this skill would generalize beyond the special starting positions I'm using, or from AI to human opponents, but even 25% of two stones in strength would be very significant. Reduce my mistakes by a factor of two. In this case that would mean reducing my mistakes by about 29 points, which is nearly the same as . Abstractly, a factor of 2 might make more sense than a fixed point-value improvement like 28, as some games are much messier than others, and I should expect that more improvement is possible in such messy games.Anyway, I have already played the next opening as white, so I will move on to writing up some joseki mistakes next.
-
hakuseki
- Dies with sente
- Posts: 100
- Joined: Thu Oct 29, 2020 6:18 am
- Rank: KGS 2 dan
- GD Posts: 0
- KGS: hakuseki
- Been thanked: 15 times
Game 35
I played moves 5–46 as white and lost 19.2 points. Most of this occurred in one corner, so I will discuss that joseki here. This game was played with group tax, which means the joseki are a bit different from those seen in more typical rule sets.
Game 35, position 1 So far, this matches the upper-left corner of the KataGo vs. KataGo game I am studying. If black had exchanged c for d before playing the hane at, then white would nobi at b. However, since black played the hane at directly, white should cut immediately at a. My move was b which lost about 0.7 points.
Game 35, position 2 It doesn't take much reading to verify that the stone is safe. White should play at a as this is the larger side. Instead I chose b which allowed black to hane at a. This was a 3.2-point mistake.
Game 35, position 3 Here I thought a looks more severe. I checked carefully that black could not capture any of my stones, and then went ahead and played a. However, black still got a strong corner and I ended up with a lot of weak points. The stronger move would have been b. This was a 2.1-point mistake.
Game 35, position 1 So far, this matches the upper-left corner of the KataGo vs. KataGo game I am studying. If black had exchanged c for d before playing the hane at
, then white would nobi at b. However, since black played the hane at directly, white should cut immediately at a. My move was b which lost about 0.7 points.Game 35, position 2 It doesn't take much reading to verify that the
stone is safe. White should play at a as this is the larger side. Instead I chose b which allowed black to hane at a. This was a 3.2-point mistake.Game 35, position 3 Here I thought a looks more severe. I checked carefully that black could not capture any of my stones, and then went ahead and played a. However, black still got a strong corner and I ended up with a lot of weak points. The stronger move would have been b. This was a 2.1-point mistake.