### What have you learned from the bots? Hex, Havannah

28 replies. Last post: 2020-09-16

What have you learned from the bots?

I was trying to understand Mootwo's playstyle, especially in the opening/early middle game.  I try to identify patterns, then it breaks them.  Maybe “Mootwo plays however it wants to play”…?

More generally, I feel I am still in the dark about opening theory, and any/all resources are most appreciated.

• lazyplayer ★ at 2020-09-14

I've learned the importance of the 4-4 points and also how to play with a13 (basically again you've to play 4-4 points but “shifted” by one diagonal). I already knew that the 4-4 opening was interesting and had to be studied but in the end I never studied it. I also knew that the diagonals “shift” but I had seen it so clearly as in the play of bots.

The fundamental principle in my view is that you've to play moves that are good locally. They must be in a favourable relationship with the closer edges and stones. Neither too close nor too far. They must also be in a favourable relationship with everything else, that is, also with the edges and stones at the opposite end of the board. But more or less the global relationship will come on their own when you're doing well locally. Anyway my main advice to you is to simply copy our openings and learn the endings before the openings. After you understand the endings then you can see that many openings lose directly and you can discard them.

• lazyplayer ★ at 2020-09-14

I've seen your profile and I've seen that you've experience in Go. Well, how can you learn in Go if a joseki makes sense or not? I'm afraid that you can't unless you play the middle-game right. I've also seen that you're already copying our openings. As a result, you're already playing openings rather well! Congrats! :D

• lazyplayer ★ at 2020-09-14

In this position you see the bots punishing you for a locally flawed stone in top-left.

The game is similar to Go. In Go the edges of the board are useful because they allow you to score points. In Hex instead connecting the edges of the board is the goal of the game. The result is that play is similar because we start from corners. After corners we may or may not play the sides of the board, it depends on the situation. If you've already the top-left and bottom-left corner, for example, and you're playing as black, then playing a stone in the left may be optimal. An example of this situation is this. The center is useful only insofar as it has enough influence on both sides. For example in this game black probably has to play the centre to keep together his stones/edges.

• lazyplayer ★ at 2020-09-14

You can think about the game as “extend the influence of both of your edges, possibly at the same time, until you win”. It also turns out that this is equivalent to “Suppress the influence of opponent's edges, possibly at the same time, until you win”. This equivalence is because we can't draw. But why we can't draw? Because locally there isn't a draw doe to the hexagonal board! You see even the global properties like no-draw are in fact deducted from the local properties.

• lazyplayer ★ at 2020-09-14

The major difference with Go is that stones do never work against us here. Your stone is always your friend. So you don't need to worry about that. On the other hand, unlike Go, a stone is really worth a ton here. If you play a really useless stone then you can resign on the spot.

• lazyplayer ★ at 2020-09-14

Generally, against c2, there are two openings that are known to be good. This one (possibly in a different order, you can play top-left first and bottom-left second). Another one is like the above except you play central cell instead of expanding the influence of your c2 (or its symmetrical equivalent). This is similar to what you were trying to do in your game, except it's done better because there are no local flaws and each stone does what it is supposed to do properly.

• lazyplayer ★ at 2020-09-14

In the game above, 8 k9 is probably the fatal error and it loses on the spot. Even top players lose on the spot like this because opening concepts are hard for everyone. One has to be specific like in the above example 9 f9 fatally wound my 8 k9 while keeping the initiative for the bot. If there was no f9 then k9 would be ok, but in fact, there is…

• lazyplayer ★ at 2020-09-14

For a13, this should be the best play. With these few examples I've already showed you the 3 main openings: C2, A13 and G3. Basically the opening is a logical consequences of many tiny details. Sometimes these tiny details conspire to make a single choice (say, the 4-4, or the centre) clearly preferable to all others and sometimes they don't.

• lazyplayer ★ at 2020-09-14

Another very solid idea is to play 11x11 games instead of 13x13 or 15x15, to get an idea of the basics (basic the more close-range interactions).

On large boards, we've very little clue on what to play, not even strong players. Even mootwo seems to play largely random at 19x19.

Lazyplayer, this is all fantastic, thank you for your detailed comments (which I am still studying).

• mouchet patrick at 2020-09-15

Lazyplayer, an elementary question from a very very weak player: in your game 2170954 with mootwo, why Black did play 3.L9 ?

• lazyplayer ★ at 2020-09-15

patrick, there is probably no difference in playing 3 L9 or 3 D5 and then 5 L9. So why L9 in general? To use the terminology above, because it's the optimal way to use extend the influence of the bottom side while making use of K12 that is already on the board. The “natural” way to play for black in that area on an empty board to play would be J9 or K10 (the symmetrical equivalents of C4 and D5) but K12 changes things there.

• lazyplayer ★ at 2020-09-15

With K12 on the board, the “obvious” ways to begin playing there in that area are L9 or K9.

You can also think about this defensively, L9 and K9 are the stones that are most blocking on the right side. But being blocking on right side is in fact the same as being connecting on the bottom because in effect when we play in that area of the board, black is usually trying to go to bottom and white is trying to go to right.

• lazyplayer ★ at 2020-09-15

In this game you see black trying to use C6 instead of B5 to extend the influence of C2. As you can see there, white immediately tried to exploit the fact that C6 needs a lot of empty space in the top area to securely connect to top. If black can play a stone there that does something else on right side then it'll be a problem for black, because it'll be forced to choose between connecting C6 to top or something else on the right side to top. This is what happened later in that game.

• lazyplayer ★ at 2020-09-15

correction, please replace black with white in “If black can play a stone there that does something”…

• lazyplayer ★ at 2020-09-15

Another great example is this. Black wins immediately while using only a tiny subset of the whole board. This is because B5 (plus C2) is so helpful on top and D10 is helpful enough on bottom. As a result of these two great supporting stones, black can play C8 and win directly on the left side.

• lazyplayer ★ at 2020-09-15

Of course “tiny subset” is an euphemism, I still need to use nearly half of the board against the best defence by white.

• lazyplayer ★ at 2020-09-15

Patrick, I've taken a look at one of your game. Look at this position. Black is losing because it can't connect bottom. To say the same thing from a defensive perspective, black is losing because it can't prevent white from connecting to right. The root of this problem is 3 I7. It's too far from bottom-right and it allows white to win there.

• lazyplayer ★ at 2020-09-15

A idea to save the game could be 9 H9 and then black could try to disconnect E9 from the left edge. He connects to right but you connect to bottom.

• lazyplayer ★ at 2020-09-15

In game #2185922, simply 5 C5 should give you a playable game at least. Your 5 G4 simply achieved nothing. White can solidify left side while threatening to connect right so basically after 5 G4 it seems already over. 11x11 is a very rapid game. Unless both plays reasonably good, the game can be over almost immediately.

• lazyplayer ★ at 2020-09-15

Anyway I've learned hex on 11x11 and i think it's the best for learning. It's so good because with some mental effort you can solve most positions.

• mouchet patrick at 2020-09-16

Lazyplayer, thank you very much for these great lessons I've now to study.

• Arek Kulczycki at 2020-09-16

On top of all that I would like to add that me and lazyplayer we disagree on 75% of ideas ;)

But he is the reigning human champion until next tournament, so I don't mean that you should distrust him!

• mouchet patrick at 2020-09-16

Don't worry. I'm probably two weak at hex to be really concerned with top level players discrepancies.

But do you disagree with the need to play much on the 11x11 version in the beginning ?

• Arek Kulczycki at 2020-09-16

No, this one I actually strongly agree. If you have such option try even 9x9, maybe vs a bot if available.

• lazyplayer ★ at 2020-09-16

We usually disagree on the “why”, but not on “what” has to be played.

• mouchet patrick at 2020-09-16

Discepancies at a deeper level therefore; with theoretical rather than practical consequences I presume.