Positional judgement | Game examples Hex, Havannah

83 replies. Last post: 2019-12-15

Reply to this topic Return to forum

Positional judgement | Game examples
  • Force majeure at 2018-12-04

    The more I try to understand hex, the more I have the impression, that playing good game consists of 3 elements:

    1) Calculating variants / brute force

    2) Strong local play based on efficient shapes etc.

    3) Positional judgement / strategic thinking

    I would like to encourage some discussion regarding 3rd point. I don’t want to go into endless details of different variants, but try to understand the direction of play, the way of thinking about hex especially after the fuseki / in the early chuban.

  • Force majeure at 2018-12-04

    Game #1

    What elements of the position do you consider crucial in planning the next move? What would be your way of thinking?

    Link to the game: http://www.trmph.com/hex/game/lg-2000947

  • lazyplayer at 2018-12-04

    Force, in reality (2) and (3) are the same at different scales. An hex game is basically a giant local battle.

  • lazyplayer at 2018-12-04

    Regarding how to think about position, you can start to try to prove one player is winning. For example, pick a reasonable white choice and try to prove it’s winning. You’ll see that the proof is elusive, and you’ll then try to amend your choice as white to look for another proof. And so on and so on. My first choice there is 12 F7, but it’s just a starting point for thinking.

  • lazyplayer at 2018-12-04

    Basically is brute force plus nice short-cuts because all variations are almost identical to each other.

  • Force majeure at 2018-12-04

    Lazy, if proving that one player wins would be so easy...I can rephrase my question and ask “how would you start proving, that one's winning a game”.

  • lazyplayer at 2018-12-04

    It’s kind of statistica, when you’ve lot of reasonable choices for each reasonable choice by opponent.

  • lazyplayer at 2018-12-04

    Ok, you can think about it like, simulations. You play a few simulated games and see winning rate.

  • blikdak at 2018-12-04

    Huh, what you just described is almost Monte Carlo tree search. Which I think is how leela works.

  • lazyplayer at 2018-12-04

    blikdak, or maybe i’ve described a “tree pruning” algorithm? in the end, all algorithms are quite the same.

  • Arek Kulczycki at 2018-12-04

    Force, I will write down my thought process after seeing the pasted above position:

    1) (5 seconds) top-left seems threatened: option = 12.d4

    2) (1 minute) is it worth it or should be sacrificed? seems threatening because it possibly cannot be surrounded by black and maybe it revives j2

    3) ok, well, for safety let’s look at other options for another minute...

    4) (1 minute) k3 and f7 are bridges and bridges are always good... but I don’t see a point of any of those moves, I mean, no immediate threat.

    5) (1 minute) e10 has merit, but how does it connect to left? let’s run a quick simulation: http://www.trmph.com/hex/board#13,a8d3i5i10e9j2e3d10k4h6b6e10b11b7c9b10c10e5

    6) not bad, let’s run a simulation for previously picked 12.d4

    7) (1 minute) http://www.trmph.com/hex/board#13,a8d3i5i10e9j2e3d10k4h6b6d4b3b4e5f7

    8) desicion: either run those simulation further or just choose one. My decision is the more interesting one: 12.e10

  • Arek Kulczycki at 2018-12-04

    I see I repeated lazyplayer’s f7 without reading before. I discarded it without any analysis, only because it doesn’t seem an immediate threat to me.

    If I were to use more time I would run simulations on f7 too, but this one seems to need like 15 minutes of analysis, I cannot focus for that long :D

  • struggler at 2018-12-04

    But what you are talking about, lazyplayer, is rather counting variants and simulations (whether with “prunning” or not). Whereas Force asked for tips and ideas on positional judgement / strategic thinking. So for me it would be an answer to the questions like “what does it mean to have a better position”, “how do you evaluate positions” (but not in terms of counting all variations till the end of game and seeing who wins).

    All in all, for other games like chess or go, there are plenty of books about every aspect of the game, positional judgement including. And in hex all we have is a few rules of thumb like “take corners”, “make multi-purpose moves” etc. Is it the characteristic of this game that it is impossible to define some general rules or just we as players are too weak to spot and describe them?

  • Arek Kulczycki at 2018-12-04

    struggler, players have ideas for rules, but their ideas differ. Nobody here is arrogant enough to write down the ideas which contradict other players ;)

  • struggler at 2018-12-04

    I have noticed it ;) and I like it, of course. However, I would greatly appreciate it if there was any source of knowledge (which would be not an opinion of a particular player but rather a fact), something like a theory about this game. And I wonder why it has not appeared yet.

  • Arek Kulczycki at 2018-12-04

    Because a fact would be something that can be proven. In hex the only proof is a brute force solution. We need opinions until some day we all just trust that leela bot is correct.

  • struggler at 2018-12-04

    Well, one could say that it applies to all strategy games. Nevertheless, for chess or go which I mentioned earlier, there were whole books on that topic way before computers (and real brute force approaches) were in use.

  • lazyplayer at 2018-12-04

    > Is it the characteristic of this game that it is impossible to define some general rules or just we as players are too weak to spot and describe them?

    We can spot them but there is no language to describe them. In chess and go they’ve invented a language to describe the most typical patterns. But there are millions more patterns that are known to strong players and have no name and no description. Most descriptions anyway are informal and they can’t be codified into precise math language.

    > I have noticed it ;) and I like it, of course. However, I would greatly appreciate it if there was any source of knowledge (which would be not an opinion of a particular player but rather a fact), something like a theory about this game. And I wonder why it has not appeared yet.

    Because (almost) nobody has written books on hex. There is one book, which used some terminology like “ladders” and “templates”, and these have come into common use. But I doubt use these concepts in my own play and I can’t recommend them. Therefore, there is nothing that I can recommend really except copying our games and running simulations and trying to understand what’s going on.

  • lazyplayer at 2018-12-04

    *But i don’t use these concepts

  • lazyplayer at 2018-12-04

    I’ve to say, I learned hex by copying Maciej. Later I’ve also copied Arek in some aspects. Arek also copied from me and Maciej. The way to mastery goes through humble copying.

  • lazyplayer at 2018-12-04

    After you’re experienced with all the typical plays that have already been invented, that is, once you’ve copied them in your games, you should start to look for variants and/or continuations. This is the only way to harass our 20 years of practical hex experience. If I win a few more games with leela (i doubt but I’ll try) then maybe i’ll write a book and you can pay me for that. :)

  • lazyplayer at 2018-12-04

    Another practical advantage of copying our initial stones is that you’ll then have to focus on the endings, and the endings are much easier to analyze. You can see what’s objectively right or wrong when the game is about to end. So not only you play better games, but you also learn more, because you stay focused on what can be learned (the ending). The beginning stones are really hard and I would say you should copy as much as possible until you’re at our level.

  • HappyHippo ★ at 2018-12-04

    I actually have written a “book” on hex that is currently in the editing phase, and I plan to put online soon (next few weeks). Of course there won’t be anything of interest to the top level players, but I think (hope) it will be of use to anyone up to around the 2050 level. I think there’s a fair amount of hex “knowledge” that fairly well known but simply hasn’t been written down anywhere and I wanted to fix that :)

    And as a bonus I’ll probably learn a thing or two when Arek et al point out all the errors ;)

  • lazyplayer at 2018-12-05

    Happy, you can share the typography you’ve used. What tools you’ve used? :)

  • psikonauta at 2018-12-05

    Wow, Hippo, that’s great, can’t wait to take a look!

  • HappyHippo ★ at 2018-12-05

    Lazy: my first attempt was at using LaTeX, I even came up with a whole “package” for typesetting diagrams, however it was extremely slow, and this became tiresome as the book got longer. Instead I ended up making the “book” a website. I have a python script I wrote which takes a marked up html file and generates diagrams as svg images. All the diagrams in the book are fully interactive so the reader can step through the sequences or try alternate variations.

  • Force majeure at 2018-12-05

    How long will be the boom HappyHippo?

  • struggler at 2018-12-05

    Wow, HappyHippo, that’s great :) can’t wait to see it published! Will it be distributed free of charge?

    And @lazyplayer – thanks for your hints, I’ll give it a try ;)

  • HappyHippo ★ at 2018-12-05

    It will be free

  • Galdian at 2018-12-06

    Great, thank you HappyHippo!

  • Force majeure at 2018-12-06

    Arek, thank you very much for your insight, that is exactly what I was looking for! To be honest, I have also considered d4 and e10, however obvious f7 was my blind spot. Finally I have chosen e11 – why would you choose e10 instead of e11?

  • Arek Kulczycki at 2018-12-06

    Force, I have no proof, but as a rule of thumb with ladders I don’t “step down”.

    Talking facts, I would probably worry about this http://www.trmph.com/hex/board#13,a8d3i5i10e9j2e3d10k4h6b6e11g10

  • Force majeure at 2018-12-07

    Game #2

    What do you think should be white’s plan? Which element of the position do you think are most important? How to utilize f1 stone?

    Link to the game: http://www.trmph.com/hex/game/lg-2000926

  • lazyplayer at 2018-12-07

    At first sight I would think that you played the correct plan at that point, but i9 is probably incorrect.

  • Arek Kulczycki at 2018-12-07

    The most important to me seems white h5 vs black i4, who plays first.

    Utilizing f1 is through c3, not g2

  • shalev at 2018-12-07

    Black looks overconnected to the bottom. There’s probably not much to be gained from battling black on the bottom (below the g9 stone), though you may scrape a bit of influence. This means white’s plan is what you did in the game: connect above the g9 stone, either between the h6 and g9 stones or else around the h6 stone. You don’t want to commit to either of these, but rather, you want to strengthen both at once.

    The top left corner is shouting out for a joseki. You played g2, which is decent, but I like c3 a bit better.

  • Force majeure at 2018-12-07

    Lazyplayer: why is that?

    Arek: When I see your comments they seem so obvious, however I couldn’t see that in the game.

  • Arek Kulczycki at 2018-12-07

    Force maybe it’s by experience. I see this sequence in an instant http://www.trmph.com/hex/board#13,f1f1j9j4c4e9h6i9g9h5g6c3 

    I wouldn’t even spend more than few moments on this, seems so forced/forcing (both)

  • Force majeure at 2018-12-07

    Arek: probably that is the reason why there is such a gap between our ratings :P Anyway, thanks for your insights, as always very helpful :)

  • Arek Kulczycki at 2018-12-07

    Force, good news is that those patterns literally repeat all the time so you will easily remember ;)

  • shalev at 2018-12-07

    To disagree a little bit, I’m not a huge fan of 10.h5, or at least, it’s not obviously the correct move. It creates a weakness on i5 which black might be able to exploit. I’d be looking at other options to see if any of them are better (maybe i3?)

  • Force majeure at 2018-12-08

    Shalev, doesn’t i3 has exactly the same weakness?

  • lazyplayer at 2018-12-08

    Why nobody likes 10 g2? Everyone ignores it? What’s problem with that? :)

  • lazyplayer at 2018-12-08

    Force, i9 usually loses to this: http://www.trmph.com/hex/board#13,f1f1j9j4c4e9h6i9j8

    it’s possible that by playing g9, probably made another error and returned the favor so to speak.

  • Arek Kulczycki at 2018-12-08

    laz, this family of things http://www.trmph.com/hex/board#13,f1f1j9j4c4e9h6i9g9g2i4j2l2k3i2i3g4h3g3h2d2

  • Force majeure at 2018-12-09

    Maybe creator of leela could arrange such position and have a deciding vote :D

  • lazyplayer at 2018-12-10

    Arek, http://www.trmph.com/hex/board#13,f1f1j9j4c4e9h6i9g9g2i4j2l2k3i2i3g4h3g3h2d2d3c3d4c5h5f6e8

  • Force majeure at 2018-12-10

    Lazy, this is very similar to options that I’ve analyzed during the game.

  • shalev at 2018-12-10

    d2 is wrong, you need b3

    Also the move order is all wrong, you should play like this, at least intuitively (I didn’t check in detail)

  • Arek Kulczycki at 2018-12-11

    @shalev, I think as white I’m delighted about black f2... I didn’t run any analysis on the h5 response, but the point is to just tenuki and live better life than 1 move before. At least in my eyes the tradeoff of g2/f2 is favourable for white.

    b3 instead of d2 is true, again, I didn’t run any analysis. Just thinking g2 in general is terrible vs a black top-left move

  • lazyplayer at 2018-12-11

    Well, I think choices like g2 are good, unless they lose immediately, heheh.

  • shalev at 2018-12-11

    I think black can afford to play f2 there. Well, actually black is probably losing overall, so I doubt I can win from that position against Arek or lazyplayer, but I think lines like this give black a decent chance (perhaps as good as he’s going to get).

  • lazyplayer at 2018-12-11

    shalev, i think you’re the most likely to win next tournament, all your modesty is very useful :)

  • Arek Kulczycki at 2018-12-11

    No offence, but leela is the most likely, no doubt :P

  • lazyplayer at 2018-12-11

    I’m willing to play some games for training purpose starting from this position: https://littlegolem.net/jsp/game/game.jsp?gid=2019562&nmove=5.

    Who is willing to play? :) Arek? Shalev? Daniel?

  • lazyplayer at 2018-12-11

    I think in that position, maybe the only chance for white is G7? What else there is? After after G7?

  • lazyplayer at 2018-12-11

    Arek, I think we human players always obsessed with spectacular tricks, but with leela, the only option is finding and playing the most principled choice.

  • lazyplayer at 2018-12-11

    After taking a look at Arek’s game vs leela:



    I would expect leela to play this vs g7:


    So this is probably what we’ve to analyze to beat the bot. :)

  • lazyplayer at 2018-12-11

    Then maybe this: http://www.trmph.com/hex/board#13,m1m1d5d10j9i4g7g3c7

    To me it doesn’t look black is hopeless. In fact I would even dare to guess that black is winning here...

  • Arek Kulczycki at 2018-12-12

    laz, in human play this pos I would say is 50%/50%. Problem is that with leela you need 60%/40% at the opening stage ;)

  • lazyplayer at 2018-12-13

    Arek, your games with leela are stuck? Feel free to ask, we’ll try to help! :P

  • Force majeure at 2018-12-14

    Game #3

    Encouraged by your discussion, I would like to show another example:

    This time I would like to understand who is ahead, and WHY. What factors would you take into account?

    Link to the game: http://www.trmph.com/hex/game/lg-2000927

  • lazyplayer at 2018-12-14

    Force, white was probably slightly ahead because you had played nothing to help you connect to top. 13 j4 would be black win i guess. I’ve also doubts on 15 c3.

  • lazyplayer at 2018-12-14

    I would also recommend to not worry too much about early play because most games are decided later anyway.

  • Arek Kulczycki at 2018-12-15

    force, according to my principles d10 is in conflict with bottom-right in a sense that black make attempts to connect bottom edge on both left and right which initially is useless.

    On the other hand white has a difficult position because all the stones are in chaos. It’s hard to put them all to use.

    I would probably prefer to play black, but it’s fairly bal;anced.

  • Arek Kulczycki at 2018-12-15

    laz, my games with leela are both lost even though I doubt they are provably lost ;)

  • Arek Kulczycki at 2018-12-15

    well, no, in fact 2019561 is clearly lost

  • lazyplayer at 2018-12-15

    Well, in 62, your stones are literally surrounded. d5 group is surrounded by white side and white stones. j9 group also surrounded by white side and stones. e7 is useless too because it’s good neithr on top nor on bottom. I would say that the “original flaw” in 62 is e7. Not sure if it’s even possible to recover from that against leela. Probably not. Leela only makes tiny errors.

  • lazyplayer at 2018-12-15

    It’s like sacrificing a pawn with no compensation and then trying to recover against a chess program. Not possible. The only difference is that here we don’t see pawns because it’s all positional and more obscure. But it should not be that obscure for you ;)

  • lazyplayer at 2018-12-15

    23 g3 instead of 23 e4 also would have been better no? you would maybe get ladder escape on top.

  • lazyplayer at 2018-12-15

    i’m discussing 61, not 62, the one you said is provably lost. i don’t know if it’s provably loss but surely it looks terrible. :D

  • Arek Kulczycki at 2018-12-15

    I had a plan for e4 and g3 but I miscalculated it. I think g3 is wrong so probably e4 is wrong too.

  • Force majeure at 2018-12-30

    Game #4

    This is probably question regarding more tactics than strategy, however I was a little bit lost here. White seems strong in the center, however there are many ajis that black may take advantage of. What would you recommend?

    Link to the game: http://www.trmph.com/hex/game/lg-2000904

  • Arek Kulczycki at 2018-12-30

    I think black is totally lost in this position. Best would maybe be 25.c6 because there are no tricks available. But 25.c6 is probably a quick white win anyway.

  • Force majeure at 2018-12-30

    Arek, I thought this is the pivotal point of the game. However, I will not argue with you – I’ll just ask when it stopped to be an equal game if you don’t mind? :)

  • lazyplayer at 2018-12-30

    Force, visually, you just have no stones on top.

  • Arek Kulczycki at 2018-12-31

    Force, I wish that position after 6.e7 is a white win because I play that way as white, but it’s pretty equal. 7.L4 I don’t like at all and possibly it’s already a turning point but then 11.k9 is a blunder and game over.

    Options for #11 are: k11 and k6. If you’re a bot then maybe also b5 but I don’t recommend :D

  • Force majeure at 2018-12-31

    Arek, thanks! Regarding 6. e7 – what is the point of this move? To be honest white shape looks strange and full of aji. Regarding 11. k9 – is it a blunder all the time (not a joseki) or in this specific position? I see that maybe k8 could be better than k9 in every possible position, but I have never seen k11 before.

  • Force majeure at 2019-01-16

    Game #5

    This is the position that I can’t figure out: