It is currently Thu Jul 07, 2022 6:48 am

All times are UTC - 8 hours [ DST ]




Post new topic Reply to topic  [ 6 posts ] 
Author Message
Offline
 Post subject: katago's aggression strategy
Post #1 Posted: Tue Nov 23, 2021 1:51 pm 
Lives in gote

Posts: 379
Liked others: 54
Was liked: 260
Rank: UK 2d Dec15
KGS: mathmo 4d
IGS: mathmo 4d
I have been studying some opening principles for katago's aggression strategy (from very conservative to very aggressive).

To me, the biggest change in strategy is how it deals with weak points and cutting points. And when splitting a side, the very conservative player will ensure life, playing away from living groups. When very conservative, they prioritise invading the opponent's moyos (approaching 3-4) and splitting their sides. When very aggressive, they slowly ensure they can prevent the opponent living easily in their area (including playing the bizzare looking 3-3 4-4 shimari), adding a move in the middle of their side and pincering any stone that approaches. This may sound contradictory but it really works.

This principle also seems to apply to cross-cut fights. The conservative player will try to rescue all the stones, saying "let's all live, you can live with more points, I'll bribe you". In fact they tend to cut to sacrifice more often. Whereas the aggressive player will choose a thin line of cutting stones to either make a wall (with lots of cutting points) by double hanes (forcefully responding to all the opponent's key point moves) or anticipate severely attacking (or even capturing) one side or the other on a large scale. The aggressive player is very sensitive to how thick they are on the rest of the board even if it is over 5 stones away. The conservative player seems to focus on local profit.

The aggressive player seems to view any side of the opponent that the opponent has played 3 moves on as small and says "I can still lean on that to build by own moyo" and after their own corners are defended, will be eager to play a very far extended moyo as long as they have many stones in that area (say six to zero) even if there are wide holes on the sides. This includes playing the 4-4, 7-3 shimari sometimes.

In a fight where the stones are one on one, (or one on zero as mentioned above - preemptive defences), this is where the differences in strategy are most apparent. Balanced play tends to defend your own group first at the boundary of access to territory/eyespace/centre influence. Conservative will do so in a more defensive way - moving towards open space rather than obsessing over the corner mutual eyespace points. Aggressive will prefer to pincer and is less likely to tenuki unless it sets up a splitting attack - the attitude seems to be "this area is already mine if I add a move to attack, so I can't let you add another move to settle."

If there is a fight where you have the advantage of 2 stones to 1 and you have to add another move to defend, then very aggressive tends to defend with the same move as the even player. 3 stones to 1 similarly - after all, you can't get around the fact you need the same 4 stones to capture a ponnuki, however much stronger you are than your opponent. However, I find that the conservative player often won't add another move to defend (because in order to be efficient and attack on a larger scale you tend to need to pincer and create another group that isn't connected to your original 2 stones) and will tenuki instead to live on the other side of the board.

If there is a fight where you have the disadvantage of 1 stone to 3 or more, this is where it gets interesting. Conservative will play to reduce the area gradually from the outside by taking big points. Conservative seems more willing to attach to the opponent's stones in order to make shape now and probe. Aggressive will leave it for now but pre-emptively get solid in the area that the opponent will profit from an attack. Then it will mess around playing all sorts of loose shape points (3rd line splits under high stones) and eventually rescue a group deep in the opponent's moyo often making a large centre wall themselves, quickly living locally, or just creating aji in sente and playing tenuki. Aggressive knows there is only one area they need in order to profit and must keep aiming to seal off that area regardless of if there are cutting points or the opponent has space to live inside.

However games between the balanced player seem to go on for a long time with a lot of tense probes and middling shapes - trying to defend slightly more efficiently but still leaving a cutting point. The timing of when the opponent exploits this and the direction they attack from is very complicated because the stones are somewhat tangled up and the opponent's stones can be vulnerable too.

My feeling is that it is easier to learn strategies from the conservative or aggressive player since the balanced AI is just too complicated (pincer - counter-pincer and on and on).

_________________
Give me triangles strong enough and I can measure the universe.

When Venus transits, we can align our clocks to one event. By measuring the angle to flat Earth at two places far apart on Earth, we can compute the distance to Venus and the Sun.


This post by dhu163 was liked by: Elom0
Top
 Profile  
 
Offline
 Post subject: Re: katago's aggression strategy
Post #2 Posted: Sat Nov 27, 2021 12:34 am 
Oza
User avatar

Posts: 2376
Location: Tokyo, Japan
Liked others: 2296
Was liked: 1325
Rank: Jp 6 dan
KGS: ez4u
Very interesting! I assume you are doing this with "playoutDoublingAdvantage". What levels are you classifying as "very conservative" and "very aggressive"?

_________________
Dave Sigaty
"Short-lived are both the praiser and the praised, and rememberer and the remembered..."
- Marcus Aurelius; Meditations, VIII 21

Top
 Profile  
 
Offline
 Post subject: Re: katago's aggression strategy
Post #3 Posted: Sat Nov 27, 2021 10:21 am 
Judan

Posts: 6719
Location: Cambridge, UK
Liked others: 436
Was liked: 3704
Rank: UK 4 dan
KGS: Uberdude 4d
OGS: Uberdude 7d
Daniel tried this out in the British Championship title match! My commentary (audio, so just the variations):


Top
 Profile  
 
Offline
 Post subject: Re: katago's aggression strategy
Post #4 Posted: Sat Nov 27, 2021 6:52 pm 
Lives in gote

Posts: 428
Liked others: 18
Was liked: 58
Rank: panda 5 dan
IGS: kvasir
It would be nice if these ideas were illustrated with diagrams. It is already a bit clearer what was meant now that I have briefly browsed the SGF, It is interesting I'll try look at it closer later.

Top
 Profile  
 
Offline
 Post subject: Re: katago's aggression strategy
Post #5 Posted: Wed Apr 27, 2022 8:22 am 
Lives in gote

Posts: 379
Liked others: 54
Was liked: 260
Rank: UK 2d Dec15
KGS: mathmo 4d
IGS: mathmo 4d
With either side, katago is confident of making life. So the main bonus that a stronger player has is noticing
the subtle changes compared to "book positions" that affect status (life and death, ability to cut) many moves before needing to read it out.

Logically, this seems to imply that strength helps most when optimising when one side is almost dead.
Where should the attacker play to finish off the opponent, or a territory/moyo that still has aji. Is a solid kill better or a more ambitious claim better?
When should the defender play activate the aji, and from which direction. Is it better to reduce on the outside first, or is it better to probe with the aji before deciding?

What does this imply for how to play with each side?
The conservative player should break up the board into lots of familiar little fights, preferring to settle large areas (get to endgame as quickly as possible, and avoid the middlegame). They say invade moyos at the last possible moment, but for conservative players, that moment comes earlier than most even if not even sente. In local fighting, if your group is alive, play more crude moves to settle the position locally and encourage the opponent to continue playing in that small area rather than making a moyo with their wall/influence. If you have the wall, balance avoiding being cut with trying to settle the position locally.

Prefer shimaris that prioritise territory over influence.

The aggressive player should invite the opponent into making a weak group in their area. Prepare to give way a little locally, just to ensure the opponent can't get eyespace in centre locally, then pincer stones that come near. If the opponent is forced to reduce lowly from further away, that remains just one weak stone and the position remains unsettled with many opportunities for them to make mistakes. If they don't enter your area, continue building it rather than invading the opponent's area, aiming to lean on their position severely for a double attack when they finally do enter your area.

If you have a cutting point that can either lead to a fight (that you are confident in simply out of aggression) or the opponent can just simplify letting you connect, then try to make the opponent overconcentrated if the opponent simplifies, while still being able to handle the fight even when the opponent has more stones on that side. Sometimes the conservative player will do the same as a probe to decide whether to defend the cut, while the even player will leave it as a fight.

Conservative: 4-4, cross fuseki, wedge safely in the opponent's moyo, gote (but at vital points)
Aggressive: 3-4, 3-3, parallel fuseki, moyo, sente (if the opponent has to take a risk to respond)

NB: some of this analysis was inspired by several pros saying something like "we see the AI line, and most of the time it makes sense and shows us clearly why a move is wrong, but the problem is that whatever line it shows us, we feel like this. Even if it showed us a crazy move, we would just nod along as AI is like God." This seems to call for more concrete analysis of what is going on, even if we can't fully appreciate pro reasoning, let alone AI, without a lot more experience.

_________________
Give me triangles strong enough and I can measure the universe.

When Venus transits, we can align our clocks to one event. By measuring the angle to flat Earth at two places far apart on Earth, we can compute the distance to Venus and the Sun.

Top
 Profile  
 
Offline
 Post subject: Re: katago's aggression strategy
Post #6 Posted: Sat May 21, 2022 10:26 am 
Lives in gote

Posts: 379
Liked others: 54
Was liked: 260
Rank: UK 2d Dec15
KGS: mathmo 4d
IGS: mathmo 4d
A few more journalistic thoughts.

Forcing moves: random or timesuji?
Many AI forcing moves are at strange timings. For example, extending from the double hane 4-4 joseki when your own important joseki stone is in atari. The numbers often show that the opponent has a difficult choice between playing the capture or responding to your threat. This seems to imply that the AI is at the level where in order to win, it strikes at the opponent's engineering as well as the game of Go. This is very similar to a timesuji that just wastes the opponent's thinking time (because they want to apply fighting spirit), but also shows just how strong the AI's judgement is. Bill mentioned the error in AI's numbers. The confidence it shows when ignoring an atari to play a forcing move is probably a measure of its strength, or else it probably wouldn't bother with such lines.

Still, it would be interesting to find a concrete case/understanding of A beats B, B beats C, C beats A showing that the AI is slacking off by finding specific weaknesses in its competition rather than improving at the game overall.




Cross fuseki (fighting games) vs Parallel fuseki (moyo games)
The general principle seems to be that in fighting shapes, there are many boundary points and all are valuable to not just territory but also life and death. And yet this also means anything can be sacrificed if there is enough profit on the other side. The main strategy seems to be to use your biggest walls to build larger areas that your opponent struggles to invade since their own groups are weak. They say don't make more than 6 groups. Invading should make your own group stronger by double attacks, which has a knock-on effect of weakening the opponent's neighbouring groups, strengthening your own on the other side and so on.

In moyo games, local boundary plays seem to be more important. This sounds inverted, and my explanation is just one perspective of a large game. However, I think this makes some sense with "the first player in an area has the advantage". We can also say that "once one vital point for an area is secured, look to the second largest vital point". In a moyo game, neither side wants to give up on the moyo, so they plan to defend their area if the opponent approaches (there is less hope of double attack in so much space unless you continue playing), so either they invade the opponent's area more deeply, or push the boundaries of their moyo.


Play the same move when far ahead/behind but not in even game?
An even game is about territory. But I have noticed weird effects in katago sometimes that mean that the same move (generally influence oriented) is chosen both when that play is far ahead/behind, but a more territorial move is chosen for an even game. Why?

My explanation is that when behind, your best chance is to build large areas and attack your opponent. When ahead, the most stable way to continue is settling as many points as possible rather than fighting. Hence, you want to get good eyespace on the side even if you give up some corner territory.

When behind, the principle seems to be to probe at weak points rather than directly cutting (as that just won't work). However, when the opponent defends, they probably defend actively, so you still have forcing moves against the cutting point. Hence, you should support such aji from as far away as you can, hoping that the opponent can't invade and double attack. The point is that you still have good local territorial aji against the weak point even if you don't have eyespace. So running is easier with a friend and let the opponent not have eyespace either if they invade.

When ahead, the principle seems to bias your own eyespace and access to all points on the board. This may mean less attacking, or at least choosing the safer direction even if not the most valuable. Normally this means a slight bias towards control of the sides rather than the corners.


local fighting choices
Using the 4-4 double pincer as a case study. Depending on direction of play, in open space we have groups A,B,C weak and adjacent.

Attacks/defense can be tighter or looser, but they might be the most valuable regardless since both are weak in open space. Normally B leans on A, creating weak points W in B's shape near A, but as C gets much weaker, it is hard for A to cut through. So C tends to run away quickly. Then B can play into the centre, making miai of curling around A or C, or lean on one side to attack the other.

In order to defend W, B sometimes leans thickly on C, threatening to kill, ensuring that even if A gets into centre, B is still connected, allowing an attack on A's eyespace. However, if such an attack is not valuable (your own stones are too weak), then you are more likely to lean on A, just tenuki, or play softer moves in the centre, aiming to make miai of controlling the centre or attacking C.

Then different things happen depending on how C continues. Tight moves might even link up to A to seal the centre, or at least force B to play a slow move to defend W, allowing C to continue developing towards the side.


Handicap games
In the opponent's area, safe eyespace is coveted more, as well as that in the "middle" of the moyo since it can develop in both directions. Generally the sides become more valuable than normal compared to the corners, but most of all, you want to settle fast and avoid the opponent having forcing moves, but even forcing moves that normally help you solidify locally can help them make more territory.

If you have any strength nearing empty space, make the most of it, developing them fast, making miai of consolidating territory/potential and getting into the opponent's area.

It is good for black to pincer to make territory from attacking since W's finds it harder to counterattack. In comparison, in a normal game, you often want to avoid attacking, but instead get endgame from weak groups.


20220626

aggressive kg likes to refuse to give an inch locally, walking into double attacks when two of its weak groups are too close to each other.

_________________
Give me triangles strong enough and I can measure the universe.

When Venus transits, we can align our clocks to one event. By measuring the angle to flat Earth at two places far apart on Earth, we can compute the distance to Venus and the Sun.

Top
 Profile  
 
Display posts from previous:  Sort by  
Post new topic Reply to topic  [ 6 posts ] 

All times are UTC - 8 hours [ DST ]


Who is online

Users browsing this forum: No registered users and 1 guest


You cannot post new topics in this forum
You cannot reply to topics in this forum
You cannot edit your posts in this forum
You cannot delete your posts in this forum
You cannot post attachments in this forum

Search for:
Jump to:  
Powered by phpBB © 2000, 2002, 2005, 2007 phpBB Group