HJJ GO Pros and Cons

Hi,
These are for G15 since I am unable to graduate :slight_smile:

My opening is always the best part of my game . 94%
The G15 have been quite hard for me. I hover at around 70%
The endgame is down to 24% through repeating the same few puzzles over and over in various combinations. I can actually solve most of them fairly easily: 24%
Cheers,
Buri


PS This is the game I just lost. As the review points out, just this one slip when I cut on the wrong side.

These are the typical small errors which are holding me back.

How small?

For me you didnā€™t read enough or played too hastily.

From both quotes, you should improve on the other side, get more focus and pleasure in reading especially in the middle game.

What about the tsumego/positional dimension? It is an indicator of your reading ability and finding shape points, you need to get 8 out of 10 right in one session to advance one step IIRC. Did you get stuck there?

Yep. That sounds about right:)

The course forced me to slow down and work harder at tumego so in that sense it has been very valuable.

This is very revealing. You didnā€™t cut on the wrong side. You didnā€™t cut at all. After the game move, white can connect all their stones, and can also capture a couple of black stones, meaning that the white group is strong, and your three black stones below might turn out to be weak. If you play on the other side, so that you actually cut, then you capture three white stones, you have no worries about your other black stones on the outside because now theyā€™re connected to everything else, and the other six white stones are weak. The difference between a strong group and a weak group is a very big deal.

So this wasnā€™t a ā€œsmall errorā€. It qualifies as a blunder: the difference between winning or losing the game. Looking at the board myself, I thought you gave away around 20 points with that one move. KataGo says more like 26 points.

Please donā€™t take this as a criticism. Iā€™ve made worse blunders in my own games. But the thing is to recognise the mistakes for what they are, and learn from them. And what a human teacher (or sharing your games in this forum) can do, which AI doesnā€™t yet do for you, is explain the nature and significance of the mistakes and help with that learning process.

If you can solve the tsumego problems but donā€™t recognise a cutting point when it comes up in your own game, then it sounds to me like youā€™re not transferring the learning from ā€œpractice mindsetā€ to ā€œperformance mindsetā€. You can see things when theyā€™re posed as a ā€œproblemā€ but arenā€™t seeing the same things during a game. Some reading material that might help with this:

Are you willing to upload full SGFs of a couple more of your games against this AI? Iā€™m curious about the AIā€™s level of play and consistency, and whether this sort of mistake is actually typical for you or whether we can see any other patterns.

2 Likes

Greetings态

I absolutely agree with what you say 100 percent. What I have learnt from these experiences is that there is almost always one error or point at which the game is lost. This was that error. As you say, the problem is lack of transfer.

The basic difference between AI and a similar level opponent is that they may well miss the error , or I can catch up by punishing theirā€™s.

Another fundamental flaw of my playing I have been made aware of is you do not stand a cat in hellā€™s chance against the AI unless you cut like crazy at every opportunity. Ironically, I have actually just started working on ā€˜Trouble Msaterā€™ by Inseong Hwang who says the following right from the start: ā€˜in cut we trust. If we were to establish a Go religion, this would undoubtedly be the central phrase.ā€™

I will take a look at the SGF thing.

Warmest regards,
Buri

But this is an artificially weakened version of the AI, right? If other 9 kyus are beating it but youā€™re not, then something else is going on here. This is another reason why Iā€™m interested to see an SGF of a full game with the AI.

ā€œCut without thinkingā€ is a habit that will get you in trouble. Itā€™s just a step above atari every time. I havenā€™t seen ā€˜Trouble Masterā€™, but I assume ā€œIn cut we trustā€ is followed by some advice on how to tell good cuts from bad ones. (Tsumego problems are supposed to help with this!)

2 Likes

Greetings,
Thank you for your generous assistance. I finally beat the AI and then beat Kate in the next game. I donā€™t know the difference between Kata and the AIā€¦ The other game was my last loss.

2024-10-11_Game_19č·Æ_166ꉋ_Kate [Lv.60]_BuriBuri [G15]_White wins by Resign.sgf (1013 Bytes)

2024-10-10_Ladder Challenge_19č·Æ_214ꉋ_Lv.202 [undefined]_BuriBuri [G15]_White wins by Resign.sgf (1.3 KB)

2024-10-10_Ladder Challenge_19č·Æ_280ꉋ_BuriBuri [G15]_Lv.202 [undefined]_White Wins (by 179 Stones.sgf (1.7 KB)

Warmest regards,
Buri

Congratulations on your wins! Does this put you up to the next level on HJJ, or do you need to score a few more points first?

I had a look at the one that you lost. (Why does it have 179 in the filename? The final score is W+4.5, right?) I can see plenty of mistakes on both sides. The AI isnā€™t as strong as you think it is. It only feels that way because youā€™re not in a positive frame of mind (and from what you described at the top of this thread, I can understand why not!)

Examples of AI mistakes in that game:

  • White 22 doesnā€™t actually capture the black stone: thereā€™s a ladder-breaker at top left!
  • White 26 is an example of Ja Choong Soo, almost a self-destructive move: next, black should push through into the centre and cut white apart.
  • White 32 is a bad time to tenuki: white should be either attacking the black group below, or defending the stones in the middle of the left side.
  • White 42 is almost a textbook example of when not to atari: should have just extended below instead.
  • White 60 was in the wrong direction: the centre is more important. Black shouldnā€™t reply locally to this move, but should play move 61 somewhere on the left or centre.

And so on. Generally the game felt like a fair fight, with black slightly ahead most of the way. It didnā€™t look like you were outclassed. In this particular game, I think the big problem wasnā€™t reading ability, it was endgame. Move 151 is typical of what was happening: you played on something that was close to a dame point. A move like this is almost the same as just passing. In the endgame, nearly every move ought to either expand your territory or reduce your opponentā€™s territory (except where youā€™re forced to defend). Other examples (although less bad) are moves 157 and 193.

From that one game, I get the feeling your results will improve if you just respect (fear?) your opponent a little bit less. Does that help at all? :slight_smile:

3 Likes

Thank you so much for your time and considerate help.

1 Like

Hello @BuriBuri , not sure if this topic is still a thing for youā€¦ just wanted to share that solving the endgame problems is much harder for someone who is used to japanese rules, becauseā€¦ HJJ GO is using chinese rules!

Thank you so much for that helpful advice. I think it will make things much easier for me.

Actually, I have been stuck for two months on one endgame problem where I am not at all clear about the values.

A6 seems to me to be the start move compared to J5 but I canā€™t get any sequence to work using either one.

Some help in counting this would be much appreciated.

Warmest regards,
Buri

Maybe something like

A6, J5,
J6, H6,
J8, F1,
E1, F9,
F4, J4,
J6

?

1 Like

Hi. Many thanks. This version seems to leave me with taking the two stones at the bottom or the two stones on the right.

The computer doesnā€™t seem to want either
Iā€™ve tried following F9 in your suggestion with:

D1 J4 J6 F4

And

J6 F4 D1 J4 A9
Regards,
Buri

Can you tell us if itā€™s black or white to play first? Is there any komi? Any prisoners? (In theory we donā€™t need to know the prisoner count ā€“ we can deduce it from the board position plus whoā€™s to play ā€“ but itā€™s a bit easier if that number is provided.)

There are five places left to play: F9, A6, J5, F4, F1. For each one, what do you think the value is? In other words, how many points is it worth, and is it sente or gote?

If the stones are already dead then you shouldnā€™t spend an extra move to take them off the board.

Black to play. Must win by 44. White by 38.
F9 Gote
A6 Gote
J5 Gote
F4 Gote 2 points
F1 Gote 6 points

I donā€™t have a clear grasp of how to calculate these values.

Best wishes,
Buri

OK, I recommend you read an endgame theory book before going further with this line of study. (And I remember earlier in this thread wondering if this level of detail on endgames is useful for you just yet.)

My current favourite is Antti Tormanenā€™s ā€œRational Endgameā€. The old Ogawa/Davies book ā€œThe Endgameā€ has a lot of good material, although you might want to skip the more advanced parts on a first reading. And thereā€™s Robert Jasiekā€™s books, if they suit your style. Or maybe thereā€™s good YouTube videos for these topics nowadays; I wouldnā€™t know.

My analysis:

A6 is gote. If black plays first, thereā€™s no territory at the top left. If white plays first, then white has 7 points at the top left (three prisoners plus one empty point). So itā€™s a 7 point swing. Older books will describe this as ā€œ7 points in goteā€. Newer books say ā€œ3.5 points by miai countingā€: for gote, you divide by 2, because the difference between black going first or white going first is 2 moves. It doesnā€™t matter too much which system you use, as long as youā€™re consistent.

By similar reasoning, F9 is 1.5 points by miai counting (3 point swing, gote) and F4 is 1 point miai (2 point swing, gote).

The other two positions are a little more complicated.

If black plays F1, then the local position is 6 points of territory (two prisoners, and two empty points at D1, E1). If white plays F1, then black will block at E1, and the two white stones are captured by snapback, so black has 5 points of territory. This is a 1 point swing. In the old style, we say 1 point, whiteā€™s sente. And in the old style, ā€œsente counts doubleā€: a 1 point sente is about as important as a 2 point gote.

For F1, the difference between black first (black plays one stone) or white first (white plays and black replies) is only 1 move not 2, so for miai counting, itā€™s still 1 point. (When itā€™s sente for one side, you donā€™t divide by 2 any more.)

For J5, you need to read a bit more. Black first is easy: black plays J5, thereā€™s no followup moves, black has 6 points at the top right. If white first, then white J5, black will block at J6, and the white J7 stone is captured because of shortage of liberties. If white J5, black J6, white H6, black J8, then white canā€™t connect at J6. So a more realistic sequence is white J5, black J6, white J4 (threatening to rescue the stone), black J8, and black now has 5 points at the top right.

So J5 is the same as F1, 1 point sente for white, or 1 point by miai counting.

Before you try solving full-board problems like this, I think youā€™ll be much better off finding a bunch of examples and practising this sort of counting until it becomes second nature. And in my opinion, thereā€™s not much value in doing this work until youā€™re at least a solid 5k. Does HJJ give you an option to skip these problems for now?

Unfortunately this is every Go game that goes to counting.

You have to solve problems like this to play a good endgame.

Usually though players can mess up, then realise A was bigger than B, and remember that for next time. Next time mess up, remember B was bigger than C and so on.

Black is starting with 37 by my count.
A6 is 4 points and allows eventual A9 for 1 more
E6 is 1 point
F1 is 1 point
F4 is 2 points
F9 is 2 points
J5 is 1 point
All of Blackā€™s moves are gote.
F1 and J5 are sente for White

I would play A6 first because itā€™s the largest. Black then has 41 points.

After that, it looks like no matter what White plays next, they share F4 and F9 for 2 points each and A9 and E6 for 1 point each which gives Black 44.

For example:
B-A6, W-F1, B-E1, W-F9, B-F4, W-A9, B-E6, W-J5, B-J6 (throw-in)

After B-E1, W-D1 is a snapback
After B-J6, W-H6 (capture), B-J8 forces W-J4, B-J6 (captures 2).