When you say things like "I played moves 112–145 as black and lost 11.7 points" what does this exactly mean? It appears to be the numbers that you report in the table but it is unclear to me if (and why) you play about 40 moves at a time, check the score evaluation and do a review before playing 40 more moves. It would appear that at move 122 it is a position from a game that you have stopped a few times to review and study the computer critique, or how does this work?hakuseki wrote: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:
Code:
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
Opening study with KataGo
-
kvasir
- Lives in sente
- Posts: 1040
- Joined: Sat Jul 28, 2012 12:29 am
- Rank: panda 5 dan
- GD Posts: 0
- IGS: kvasir
- Has thanked: 25 times
- Been thanked: 187 times
Re: Opening study with KataGo
-
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 am following the study plan I outlined in this post: viewtopic.php?p=272857&sid=113f380b0098 ... 2c#p272857kvasir wrote:When you say things like "I played moves 112–145 as black and lost 11.7 points" what does this exactly mean? It appears to be the numbers that you report in the table but it is unclear to me if (and why) you play about 40 moves at a time, check the score evaluation and do a review before playing 40 more moves. It would appear that at move 122 it is a position from a game that you have stopped a few times to review and study the computer critique, or how does this work?
-
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
Sorry, I should clarify this a bit more. When I review my game in KaTrain, it shows a "point loss" for each move. For example, in my review I might mention that a move is a "2.1-point mistake." This is based on KataGo's evaluation as displayed in KaTrain.kvasir wrote:When you say things like "I played moves 112–145 as black and lost 11.7 points" what does this exactly mean?
My total point loss is calculated by adding up the loss for every single move. Sometimes a move may have a positive score, which will reduce the point loss slightly. However, most of my moves get a negative score and increase the point loss.
As this implies, the opponent's mistakes have no direct effect on my point loss. So I can lose many points and still win the game, provided my opponent is losing even more points. However, for study purposes, I do not care about winning or losing. I only care about minimizing my point loss.
-
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
By the way, I've finished my first playthrough as white now. My total loss was 78.7 points. This is worse than my result as black. I wonder if I'm having a bad day? Or perhaps it is just random variation. Anyway, my average is now 68.4 points lost.
Based on these results, I think I will set two goals. My true goal is to have a total loss of 29 points or less; my fallback goal is to have a total loss of 40 points or less.
The true goal is half of my better result out of my first two games; the fallback goal is about 28 points better than my average result. Regardless of whether I reach either goal, I may decide to continue studying this game as long as I am seeing good progress, or terminate the study early if I am not making progress.
I said I will focus on one section of the game at a time. Overall, I made the most mistakes in the midgame section covering moves 81–111, so starting there might make the most sense. However, the difference between sections is actually not so big, and I think it might be more exciting to start with the endgame and work my way back towards the beginning. Perhaps that is what I'll try.
Based on these results, I think I will set two goals. My true goal is to have a total loss of 29 points or less; my fallback goal is to have a total loss of 40 points or less.
The true goal is half of my better result out of my first two games; the fallback goal is about 28 points better than my average result. Regardless of whether I reach either goal, I may decide to continue studying this game as long as I am seeing good progress, or terminate the study early if I am not making progress.
I said I will focus on one section of the game at a time. Overall, I made the most mistakes in the midgame section covering moves 81–111, so starting there might make the most sense. However, the difference between sections is actually not so big, and I think it might be more exciting to start with the endgame and work my way back towards the beginning. Perhaps that is what I'll try.
-
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
Today I've been playing some late endgames.
Although I managed a positive score the first couple times, this has not been the pattern generally. Thankfully I have not made major blunders (except once); however, often my opponent makes some mistake that I fail to exploit, or exploit only partially.
Normally I don't expect my AI opponent to make a lot of mistakes, but the late endgame actually seems to be a bit of a weak point for the Policy Weighted AI in KaTrain.
I think this can be good training for me. I need to always be vigilant for my opponent's mistakes, and extract every possible point from them when they occur.
Although I managed a positive score the first couple times, this has not been the pattern generally. Thankfully I have not made major blunders (except once); however, often my opponent makes some mistake that I fail to exploit, or exploit only partially.
Normally I don't expect my AI opponent to make a lot of mistakes, but the late endgame actually seems to be a bit of a weak point for the Policy Weighted AI in KaTrain.
I think this can be good training for me. I need to always be vigilant for my opponent's mistakes, and extract every possible point from them when they occur.
-
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 36
Here's my largest late endgame mistake so far.
Game 36, position 1 After
, white's lower right corner looks a bit thin. However, I read one sequence, and decided black couldn't do anything, so although still slightly nervous, I played 
. Here's the sequence I read:
Black doesn't have enough liberties to capture white's one-eye group. White can even ignore
. So I thought this was okay.
Here's black's actual attack:
White cannot save the entire corner now. At least 23 points are lost. I lost even more in the actual game.
The misread is one thing, but I am also interested in the bad thinking habit this surfaces. When I played in the first diagram I did so with some nervousness. I think subjectively, I was only about 70% confident that my lower right group was safe.
Actually, in that kind of situation I should keep reading out more sequences. But I also think that, at the end of all my reading, if I still think there is only a 70% chance that my group is safe, I should play a defensive move. I should use the concept of "expected value" to play the subjectively best move in my estimation, rather than simply playing the move I judge most likely to be optimal.
This was not always my approach to Go, but I have come over to this view within the last couple years. However, I still have many bad habits.
Game 36, position 1 After
, white's lower right corner looks a bit thin. However, I read one sequence, and decided black couldn't do anything, so although still slightly nervous, I played . Here's the sequence I read:Black doesn't have enough liberties to capture white's one-eye group. White can even ignore
. So I thought this was okay.Here's black's actual attack:
White cannot save the entire corner now. At least 23 points are lost. I lost even more in the actual game.
The misread is one thing, but I am also interested in the bad thinking habit this surfaces. When I played
in the first diagram I did so with some nervousness. I think subjectively, I was only about 70% confident that my lower right group was safe.Actually, in that kind of situation I should keep reading out more sequences. But I also think that, at the end of all my reading, if I still think there is only a 70% chance that my group is safe, I should play a defensive move. I should use the concept of "expected value" to play the subjectively best move in my estimation, rather than simply playing the move I judge most likely to be optimal.
This was not always my approach to Go, but I have come over to this view within the last couple years. However, I still have many bad habits.
-
kvasir
- Lives in sente
- Posts: 1040
- Joined: Sat Jul 28, 2012 12:29 am
- Rank: panda 5 dan
- GD Posts: 0
- IGS: kvasir
- Has thanked: 25 times
- Been thanked: 187 times
Re: Opening study with KataGo
OK, I get it. My other question is why start from different positions in the same game, it seems to risk that you get familiar with that game first and then play from the positions with a better idea how to play. Anyway, it is an interesting approach to play certain number of moves from the same position and repeat until you get a hang of it. I think I might try that myself.hakuseki wrote:Sorry, I should clarify this a bit more. When I review my game in KaTrain, it shows a "point loss" for each move. For example, in my review I might mention that a move is a "2.1-point mistake." This is based on KataGo's evaluation as displayed in KaTrain.
BTW it might be easier (or more interesting) to follow your thread if you can post the SGFs or diagrams showing how the playout actually went. I find myself confused by the half- and quarter-board diagrams because many things are playable in isolation. If you don't want to share I understand that too, playing the computer can be embarrassing
-
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
Followup results
I've played the equivalent of 10 games now, starting from various positions of this kifu I'm studying. And so far I've seen no net improvement! My average score has actually gotten worse, but only by 0.037, which I guess is not significant.
Here are my updated average statistics:
Based on these statistics, I guess the opening is as good a focus as any for now. I think I will need a more concrete plan to improve, as just playing games and reviewing has not been enough.
I think I will try coming up with some kind of checklist or flowchart to follow in the opening. Then I will actually write down my notes for each move as I play it, to make sure I am actually following the steps.
Perhaps I can develop this flowchart by re-examining these last 10 openings. I will ask myself what reasoning steps I could have taken to avoid each mistake. Then I will edit the most commonly recurring steps down into a flowchart.
Here are my updated average statistics:
Code: Select all
Stage Moves Loss
initial moves 1–4 N/A
opening 5–46 13.53
late opening/early midgame 47–80 13.41
midgame 81–111 13.33
ōyose 112–145 9.87
early endgame 146–174 7.24
endgame 175–210 5.11
late endgame and dame 211–243 5.95
TOTAL 1-243 68.44
I think I will try coming up with some kind of checklist or flowchart to follow in the opening. Then I will actually write down my notes for each move as I play it, to make sure I am actually following the steps.
Perhaps I can develop this flowchart by re-examining these last 10 openings. I will ask myself what reasoning steps I could have taken to avoid each mistake. Then I will edit the most commonly recurring steps down into a flowchart.
-
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
Checklist followup
Good news: I made a checklist, then I played ten more openings. This time my average point loss was 11.03 points, which is an improvement of 2.5 points!
Bad news: I didn't actually use my checklist very much. It's just too long and annoying. Here it is, for the curious:
(note: the forum somehow messed up the indentation on this, even though it's inside a code block)
I do think the process of making the checklist was instructive, at least. And I did use it a little bit.
Basically, in the ten openings I analyzed to produce this checklist (from the cross-hoshi starting position), I found that a lot of my mistakes had to do with fighting, especially with defense. Joseki mistakes were less of an issue.
In my newer set of 10, I am thankfully seeing slightly fewer mistakes overall. Reading during fights is still an issue, and sente seems like an issue. I also seem to have more joseki mistakes this time. I am now planning to repeat my analysis and produce a new checklist, and also some joseki diagrams.
Bad news: I didn't actually use my checklist very much. It's just too long and annoying. Here it is, for the curious:
(note: the forum somehow messed up the indentation on this, even though it's inside a code block)
Code: Select all
1. Review the opponent's move.
-> did they ignore a peep or cutting point?
-> think about cutting. Read variations.
-> did they extend on the 4th line?
-> think about knight's move approach
2. If it were my opponent's turn now, what moves would they consider?
3. Decide what move I want to play. Use lots of intuition and reading.
4. Review my move idea.
-> Is it a big move, an urgent or fighting move, or a joseki move?
-> big move
-> Am I sure there's no urgent move?
-> Are there any contacting stones or cutting points?
-> Are there any weak groups?
-> What do I think the coin value is?
-> fighting move
-> am I attacking or defending?
-> attacking
1 is this aji keshi?
2 am I actually weakening an important group?
3 can I attack from a different direction?
4 am I surrounding? if so, can I make my opponent crawl instead?
5 am I making good shape or bad shape?
-> defending
1 could my group be safe already?
2 is this actually big enough to defend?
3 can I fight back instead?
4 am I making my group heavy?
-> can I sacrifice it?
5 am I between two enemy groups?
-> can I kosumi out?
6 can I make good shape?
-> nozoki?
-> tsuke?
1 are there other fights on the board?
-> does this fight involve more weak stones?
2 did I read at least 3 moves deep?
3 did I read all the obvious variations?
4 did I find a move that's almost as good?
-> have I searched for other options similar to that move?
-> joseki move
-> do I know this joseki well?
-> no
-> is this a fight? If so, go to fighting moves
-> read several possibilities and variations at least 3 moves deep
-> Is this move multipurpose?
5. Play my move.
Basically, in the ten openings I analyzed to produce this checklist (from the cross-hoshi starting position), I found that a lot of my mistakes had to do with fighting, especially with defense. Joseki mistakes were less of an issue.
In my newer set of 10, I am thankfully seeing slightly fewer mistakes overall. Reading during fights is still an issue, and sente seems like an issue. I also seem to have more joseki mistakes this time. I am now planning to repeat my analysis and produce a new checklist, and also some joseki diagrams.
-
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
Joseki notes for games 37–46: Double Approach
Joseki notes for games 37–46: double approach
(Note: these games were played with group tax, which affects opening strategy somewhat)
4-4 point, double approach, attachment If white plays a next, then black should check the ladder. If it works, then cutting at b is the best move. Otherwise, a move at c, d, or e is good enough.
However, in my game 37, white played elsewhere on the board instead of at a. In this case, blocking the corner with e is the best move, about 0.8 points better than cutting.
4-4 point double approach, knight's move
seems strange, but KaTrain's Policy Weighted AI has been playing this move against me often.
When the board is mostly empty, an extension such as a–e seems to be favored. The 3-3 invasion at f is evaluated at -0.56 points.
In other situations, e.g. where the sides are more crowded, the 3-3 invasion may be good. In game 37 I played the invasion although it was not favored. Let's look at the continuation: First, let's look at c. I actually read this sequence out before playing and thought this result was OK. It was a failure of positional judgment. lost 3.4 points and 
lost 1 point (compared to d).
Black should instead play. Now white may tenuki or may play some side move like . at b is also acceptable and white's response will be similar.
4-4 point keima and ōgeima double approach At least in the context of my game 37, b was the only good option for black. It is also a good move against the approach, with or without the presence of 
.
The other marked points varied in evaluation from about -0.6 to -0.4. Of these, a and c seem the worst yet have high policy weight.
(Note: these games were played with group tax, which affects opening strategy somewhat)
4-4 point, double approach, attachment If white plays a next, then black should check the ladder. If it works, then cutting at b is the best move. Otherwise, a move at c, d, or e is good enough.
However, in my game 37, white played elsewhere on the board instead of at a. In this case, blocking the corner with e is the best move, about 0.8 points better than cutting.
4-4 point double approach, knight's move
seems strange, but KaTrain's Policy Weighted AI has been playing this move against me often.When the board is mostly empty, an extension such as a–e seems to be favored. The 3-3 invasion at f is evaluated at -0.56 points.
In other situations, e.g. where the sides are more crowded, the 3-3 invasion may be good. In game 37 I played the invasion although it was not favored. Let's look at the continuation: First, let's look at c. I actually read this sequence out before playing
and thought this result was OK. It was a failure of positional judgment. lost 3.4 points and lost 1 point (compared to d).Black should instead play
. Now white may tenuki or may play some side move like . at b is also acceptable and white's response will be similar.4-4 point keima and ōgeima double approach At least in the context of my game 37, b was the only good option for black. It is also a good move against the
approach, with or without the presence of .The other marked points varied in evaluation from about -0.6 to -0.4. Of these, a and c seem the worst yet have high policy weight.
-
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
Additional Joseki notes for games 37-46
Additional Joseki notes for games 37-46
(Note: these games were played with group tax, which affects opening strategy somewhat)
4-4 point direct 3-3 invasion There are many ways of dealing with a 3-3 invasion, but in a cross-hoshi fuseki with group tax, the knight's move at
is favored by about 0.2 points. The sequence through 
is expected; 
may either be as shown or at a.
Actually, the direct 3-3 invasion rarely seems to come up in these games, but I think I might see it if I ever play group tax games against opponents more familiar with modern Go, so this seems worth knowing.
4-4 point, keima approach, side hoshi pincer, tenuki Kosumi-tsuke is a famous handicap joseki; when the supporting stone at is present, a follow-up at a is considered superior to other responses such as b or c.
In this situation in game 41, I played a move elsewhere on the board. But the priority of a is quite high, almost equivalent to playing in an empty corner. So black should play at a.
4-4 point, keima approach supported by side hoshi, side invasion Black invades at
, so naturally white covers at . If black plays elsewhere, what is white's local continuation?
I sort of assumed that making the tiger's mouth at c must be the ideal shape. But b and especially a turned out to be slighly better, at least in the context of my game 42. I will definitely consider these moves in a similar situation, and not just assume that c is the move.
4-4 point, keima approach, attachment Black should continue at a (linking up all stones), or b (possibly sacrificing the corner, but building side influence). Black c was my greedy mistake in game 43.
4-4 point keima approach ōgeima response continuation This sequence is becoming quite familiar to me. Usually black will continue at a, or possibly play an extension such as b.
If black plays elsewhere instead, then white may continue at a. This seems to be more urgent than approaching a 4-4 stone or defending a 4-4 stone, but less urgent than answering a double approach or 3-3 invasion. I tried setting up a position with a 3-4 stone in another corner, and a was considered very slightly less urgent than approaching the 3-4 stone.
Perhaps this seems like a rather fine point to be bringing up, but this sequence really is appearing a lot in my games with group tax, so it is valuable for me to know the priority of this move.
4-4 point, keima approach, ōgeima extension, one-space approach A move such as a, b, c, or d would be good. I thought these moves seemed slow and played e, but this was suboptimal.
EDIT: Another interesting move is f. This move happens to appear in Josekipedia. Although the policy model rates this move very low, it only loses 0.20 points compared to a (in the context of my game 46). Furthermore, it is actually 0.71 points better than a in the context of a (somewhat different) position in my game 48, although is on the fourth line in that game.
Shape point White is running towards the center, and would like to run with good shape. My move was g, but this is aji keshi. I thought about a, but it felt to me like an incomplete shape.
In fact,
is a good move, as either a or b would complete a good shape.
(Note: these games were played with group tax, which affects opening strategy somewhat)
4-4 point direct 3-3 invasion There are many ways of dealing with a 3-3 invasion, but in a cross-hoshi fuseki with group tax, the knight's move at
is favored by about 0.2 points. The sequence through is expected; may either be as shown or at a.Actually, the direct 3-3 invasion rarely seems to come up in these games, but I think I might see it if I ever play group tax games against opponents more familiar with modern Go, so this seems worth knowing.
4-4 point, keima approach, side hoshi pincer, tenuki Kosumi-tsuke is a famous handicap joseki; when the supporting stone at
is present, a follow-up at a is considered superior to other responses such as b or c.In this situation in game 41, I played a move elsewhere on the board. But the priority of a is quite high, almost equivalent to playing in an empty corner. So black should play at a.
4-4 point, keima approach supported by side hoshi, side invasion Black invades at
, so naturally white covers at . If black plays elsewhere, what is white's local continuation?I sort of assumed that making the tiger's mouth at c must be the ideal shape. But b and especially a turned out to be slighly better, at least in the context of my game 42. I will definitely consider these moves in a similar situation, and not just assume that c is the move.
4-4 point, keima approach, attachment Black should continue at a (linking up all stones), or b (possibly sacrificing the corner, but building side influence). Black c was my greedy mistake in game 43.
4-4 point keima approach ōgeima response continuation This sequence is becoming quite familiar to me. Usually black will continue at a, or possibly play an extension such as b.
If black plays elsewhere instead, then white may continue at a. This seems to be more urgent than approaching a 4-4 stone or defending a 4-4 stone, but less urgent than answering a double approach or 3-3 invasion. I tried setting up a position with a 3-4 stone in another corner, and a was considered very slightly less urgent than approaching the 3-4 stone.
Perhaps this seems like a rather fine point to be bringing up, but this sequence really is appearing a lot in my games with group tax, so it is valuable for me to know the priority of this move.
4-4 point, keima approach, ōgeima extension, one-space approach A move such as a, b, c, or d would be good. I thought these moves seemed slow and played e, but this was suboptimal.
EDIT: Another interesting move is f. This move happens to appear in Josekipedia. Although the policy model rates this move very low, it only loses 0.20 points compared to a (in the context of my game 46). Furthermore, it is actually 0.71 points better than a in the context of a (somewhat different) position in my game 48, although
is on the fourth line in that game.Shape point White is running towards the center, and would like to run with good shape. My move was g, but this is aji keshi. I thought about a, but it felt to me like an incomplete shape.
In fact,
is a good move, as either a or b would complete a good shape.
Last edited by hakuseki on Sat Jun 25, 2022 6:05 am, edited 1 time in total.
-
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
Second checklist
Here's my second attempt at a checklist.
Code: Select all
Global concerns:
What is the second-biggest-seeming area?
Could there be a bigger move there that I haven't checked yet?
Reading:
Did I think of every possible local move?
What sequences can follow from each move?
Which of those outcomes is best?
If I think my move is sente:
Am I sure that my followup actually works?
Am I sure that my opponent's defense can't be sente?
Extra questions about defense:
Do I actually need to defend?
Which move minimizes aji?
What are the endgame followups?
-
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
Ten more openings
I've played ten more openings, and this time I achieved an average point loss of 8.87 points, which is more than 2 points better than my previous result.
This progress gives me hope that I may gain a stone in strength this year.
I'll post a more detailed analysis soon.
This progress gives me hope that I may gain a stone in strength this year.
I'll post a more detailed analysis soon.
-
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: Ten more openings
I mentioned my mistakes totaled about 8.87 points per opening in this last set.
Of these, about 1.1 points were lost to joseki mistakes; about 1.5 points to direction of play mistakes, and about 2.2 points to technique mistakes.
"Technique mistakes" is a sort of vague category, but I am referring to questions like: Should I make a solid or hanging connection? Should I push from behind or run ahead with a keima? Should I hane or extend? Generally, these situations did not involve immediate life and death concerns, so the focus is more on judging shape qualities in a particular context.
There were other mistakes, but these seem sufficient to focus on for now. I will make some more detailed posts with diagrams soon.
I also categorized some moves as "dogma mistakes," by which I mean moves that humans would probably consider reasonable but that don't follow the AI style perfectly. These mistakes were very small, but there were a lot of them, so I wanted to see whether they're worth paying attention to. However, they only totaled about 0.4 points per opening, so I will set this concern aside for now.
Of these, about 1.1 points were lost to joseki mistakes; about 1.5 points to direction of play mistakes, and about 2.2 points to technique mistakes.
"Technique mistakes" is a sort of vague category, but I am referring to questions like: Should I make a solid or hanging connection? Should I push from behind or run ahead with a keima? Should I hane or extend? Generally, these situations did not involve immediate life and death concerns, so the focus is more on judging shape qualities in a particular context.
There were other mistakes, but these seem sufficient to focus on for now. I will make some more detailed posts with diagrams soon.
I also categorized some moves as "dogma mistakes," by which I mean moves that humans would probably consider reasonable but that don't follow the AI style perfectly. These mistakes were very small, but there were a lot of them, so I wanted to see whether they're worth paying attention to. However, they only totaled about 0.4 points per opening, so I will set this concern aside for now.
-
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
Joseki notes for games 47-56
Joseki notes for games 47-56
(Note: these games were played with group tax, which affects opening strategy somewhat)
4-4 point, keima approach, keima extension, contact play This is very common AI joseki, but sometimes I've forgotten how it ends. White will play a, and then what?
I played b but this is too passive.
should either be at a or tenuki. Let's look at both:
White will probably cut at, but this isn't scary for black. Black is happy to just keep pushing like this. There are some other variations, but black's goal will be the same: build a wall.
If black tenukis, white may double hane, but black can simply tenuki again. The black stones will are alive via the miai of a and b.
4-4 point, low approach, kosumi-tsuke transposition Various move orders; may result in this shape in the upper-left; this particular order is from my game 47.
Where should black play next? I played a, which I have seen in handicap games. This move would have been correct if the marked stone (the upper right star point) were white, but stronger moves are available in this position. Black should pincer the
stone, for instance with b, c, or d.
4-4 point, low approach, one-space low pincer Here's what I've noticed about how KataGo responds to pincers when playing with group tax:
Generally, white will not invade at the 3-3 point, in contrast to games without group tax, in which case 3-3 invasions are quite common.
In the case of a near pincer (whether on the third or fourth line), the one-space jump at a is good.
For more distant pincers, a double approach or counter-pincer will usually be preferred.
In any case, answering a pincer does not seem be urgent. Tenuki, e.g. to approach another corner, is often the best move.
In my game 48, I played a, but b would have been about 0.4 points better. I also misplayed the continuation, as shown below:
4-4 point, low approach, one-space low pincer, double approach, continuation Even after, the local situation is not urgent; white would most likely tenuki, then come back and play later, or perhaps directly invade at instead. But I think the sequence shown in the diagram is easy to understand. In my game 48, I played the atari at a instead of , which was a mistake of 2.5 points.
4-4 point, low approach, ōgeima, ōgeima, variation 1 In my previous games,
has been at e, but the move shown here is probably more common. In response, I played 
at e, which is not terrible, but the other options shown here are about half a point better.
Black may aim for the corner with a, in which case white will turn at b; otherwise, black may aim to press white down with b, c, or d.
4-4 point, low approach, ōgeima, ōgeima, variation 2 Here's another variation.
is a mistake of about 0.5 points. should be at a. I played at b, but this was a mistake of about 2 points.
4-4 point, double approach, contact play, hane, extend, push I find this rather characteristic of the group tax opening style. Continuation follows.
is not really sente. In my game, I responded at a. The Josekipedia move is b. Both of these are inferior to c. Tenuki is also a good option.
In the case of tenuki, b and c may be considered miai. If black reduces the corner with b, white moves to the side with c. If black instead plays c, or some other move to secure the side, then white takes the corner with b.
4-4 point, approach, low pincer, one-space jump, two-space jump This sequence may favor black slightly. Perhaps should have been at a. Continuation follows:
Black has several good responses to. I played e, but this lukewarm move is not great. The other labeled moves are good.
I'm especially interested in b. This kind of slide under the 4-4 point has fallen off my radar along with the old "4-4 point, keima approach, keima extension, slide" joseki. However, in the context of this pincer fight, it seems to be effective. I've also seen it in other, similar contexts. I'll try to keep it in mind from now on.
4-4 point, approach, two-space high extension, 3-3 invasion In this position, at a is favored over b by about 0.2 points. I played b because I thought it limited white's movement more and the presence of the stone would help reinforce black's weakness at a. But this cutting point did end up becoming a problem later in my game 51.
4-4 point, double approach, contact play, hane, extend Here are some of white's various options for, but I'm mainly interested in discussing a and b. When should white choose a and when should white choose b?
I usually play a because I feel it gives me slightly more territory than b. But recently I feel that b damages the stone less than a. So I should decide whether I want to establish a group on the top side; if so, then I should lean more towards playing b.
4-4 point, inside contact play, inside hane, extend, push Here's a sequence from my game 56. I played as white here. In the context of a mostly-empty board, seems a bit premature. For , either the inside or outside hane is fine; I've also seen other whole-board situations where the outside hane was preferred. I played because it felt to me like a move AI would play; this was correct, but I did not really understand the meaning of this move and I made mistakes in the continuation.
In particular, I think if the hane at a were a good move for black, this would suggest that is bad. However, black should not hane because the cut at b would be severe.
Since black cannot hane, black extends with. My push at was a mistake, because this time black can hane at and the cut is less severe. Instead of continuing to push, white should have defended the corner with a move such at a or b.
The mistakes discussed above accounted for an average of 1 lost point per game within this set of ten openings.
(Note: these games were played with group tax, which affects opening strategy somewhat)
4-4 point, keima approach, keima extension, contact play This is very common AI joseki, but sometimes I've forgotten how it ends. White will play a, and then what?
I played b but this is too passive.
should either be at a or tenuki. Let's look at both:White will probably cut at
, but this isn't scary for black. Black is happy to just keep pushing like this. There are some other variations, but black's goal will be the same: build a wall.If black tenukis, white may double hane, but black can simply tenuki again. The black stones will are alive via the miai of a and b.
4-4 point, low approach, kosumi-tsuke transposition Various move orders; may result in this shape in the upper-left; this particular order is from my game 47.
Where should black play next? I played a, which I have seen in handicap games. This move would have been correct if the marked stone (the upper right star point) were white, but stronger moves are available in this position. Black should pincer the
stone, for instance with b, c, or d.4-4 point, low approach, one-space low pincer Here's what I've noticed about how KataGo responds to pincers when playing with group tax:
Generally, white will not invade at the 3-3 point, in contrast to games without group tax, in which case 3-3 invasions are quite common.
In the case of a near pincer (whether on the third or fourth line), the one-space jump at a is good.
For more distant pincers, a double approach or counter-pincer will usually be preferred.
In any case, answering a pincer does not seem be urgent. Tenuki, e.g. to approach another corner, is often the best move.
In my game 48, I played a, but b would have been about 0.4 points better. I also misplayed the continuation, as shown below:
4-4 point, low approach, one-space low pincer, double approach, continuation Even after
, the local situation is not urgent; white would most likely tenuki, then come back and play later, or perhaps directly invade at instead. But I think the sequence shown in the diagram is easy to understand. In my game 48, I played the atari at a instead of , which was a mistake of 2.5 points.4-4 point, low approach, ōgeima, ōgeima, variation 1 In my previous games,
has been at e, but the move shown here is probably more common. In response, I played at e, which is not terrible, but the other options shown here are about half a point better.Black may aim for the corner with a, in which case white will turn at b; otherwise, black may aim to press white down with b, c, or d.
4-4 point, low approach, ōgeima, ōgeima, variation 2 Here's another variation.
is a mistake of about 0.5 points. should be at a. I played at b, but this was a mistake of about 2 points.4-4 point, double approach, contact play, hane, extend, push I find this
rather characteristic of the group tax opening style. Continuation follows. is not really sente. In my game, I responded at a. The Josekipedia move is b. Both of these are inferior to c. Tenuki is also a good option.In the case of tenuki, b and c may be considered miai. If black reduces the corner with b, white moves to the side with c. If black instead plays c, or some other move to secure the side, then white takes the corner with b.
4-4 point, approach, low pincer, one-space jump, two-space jump This sequence may favor black slightly. Perhaps
should have been at a. Continuation follows:Black has several good responses to
. I played e, but this lukewarm move is not great. The other labeled moves are good.I'm especially interested in b. This kind of slide under the 4-4 point has fallen off my radar along with the old "4-4 point, keima approach, keima extension, slide" joseki. However, in the context of this pincer fight, it seems to be effective. I've also seen it in other, similar contexts. I'll try to keep it in mind from now on.
4-4 point, approach, two-space high extension, 3-3 invasion In this position,
at a is favored over b by about 0.2 points. I played b because I thought it limited white's movement more and the presence of the stone would help reinforce black's weakness at a. But this cutting point did end up becoming a problem later in my game 51.4-4 point, double approach, contact play, hane, extend Here are some of white's various options for
, but I'm mainly interested in discussing a and b. When should white choose a and when should white choose b?I usually play a because I feel it gives me slightly more territory than b. But recently I feel that b damages the
stone less than a. So I should decide whether I want to establish a group on the top side; if so, then I should lean more towards playing b.4-4 point, inside contact play, inside hane, extend, push Here's a sequence from my game 56. I played as white here. In the context of a mostly-empty board,
seems a bit premature. For , either the inside or outside hane is fine; I've also seen other whole-board situations where the outside hane was preferred. I played because it felt to me like a move AI would play; this was correct, but I did not really understand the meaning of this move and I made mistakes in the continuation.In particular, I think if the hane at a were a good move for black, this would suggest that
is bad. However, black should not hane because the cut at b would be severe. Since black cannot hane, black extends with
. My push at was a mistake, because this time black can hane at and the cut is less severe. Instead of continuing to push, white should have defended the corner with a move such at a or b.The mistakes discussed above accounted for an average of 1 lost point per game within this set of ten openings.