Hex bot battle continued Hex, Havannah

166 replies. Last post: 2020-04-25

Reply to this topic Return to forum

Hex bot battle continued
  • Bill LeBoeuf ★ at 2019-09-24

    Wow, Leela got beat by Moo in a game where Leela had the swap advantage (game 2122088).  Very amazing.

  • Andrey at 2019-09-25

    Will be nice for mooone/two owner to introduce himself and to know smth. about these probably AI's.

  • Arek Kulczycki at 2019-09-25

    Hooray, mootwo confirms that I'm correct claiming that this is black win ;)

  • lazyplayer ★ at 2019-09-25

    Arek, hahah :D

  • lazyplayer ★ at 2019-09-25

    Arek, but indeed the game seems very principled and it could even be considered a refutation of the double 4-4 opening!

    If you want to try 5-5 opening, then maybe try again this: https://littlegolem.net/jsp/game/game.jsp?gid=2090087&nmove=7

  • mootwo at 2019-09-25

    Excuse me for skipping introduction, the bot will be introduced later at some point.

    In 2122088, the bot thought it was behind until move 22.g10, after that it considered itself significantly ahead. As white, it would play g7. Seems like a blunder by Leela?

  • Arek Kulczycki at 2019-09-25

    22.g7 seems a natural human choice, but leela might have been right to look for complications.

    This is a real threat

  • lazyplayer ★ at 2019-09-26

    mootwo, is the bot MCTS-based or is it based on some minimax variant?

  • mootwo at 2019-09-26

    I let it play against itself for a few moves link , and it does change its mind after 27.e8 is played to black being ahead, and in the end black would win this game.

  • mootwo at 2019-09-26

    Sorry, fixed link

  • lazyplayer ★ at 2019-09-26

    mootwo, take a look at this, maybe is it an improvement on white play?

  • mootwo at 2019-09-26

    The bot seems to like that slightly less. Black still won when I let it play against itself from that spot: link

  • lazyplayer ★ at 2019-09-26

    Well, 5 L6 is crazy and 6 I3 can't be the right reply… somehow bots win vs humans without understanding anything for real… :D

  • Arek Kulczycki at 2019-09-28

    Oh, so indeed leela was correct. But it only found another way to lose! :)

    @lazy, in fact both L6 and i3 are reaaonable, they occur in most games of this kind but later :) From experience we know that in games of similiar nature there are key moves and sometimes indeed we should play them earlier than we like.

    On the other hand L6 is dependant on L9 so I dont see the point of playing L6 first :D

  • lguser at 2019-10-04

    LeelaBot loses again to MooTwo: http://www.littlegolem.net/jsp/game/game.jsp?gid=2123807

    this game had an unorthodox opening with MooTwo's h8 response to the first stone at k7

  • lazyplayer ★ at 2019-10-04

    MooTwo, congrats, it seems the initial stones are really hard even for the bots.

  • lazyplayer ★ at 2019-10-04

    The irony of hex is that intial stones seems impossibly hard but then a few moves later it's game over… so there has to be a way to play right even at the beginning but it's just a little too hard for us… maybe these MCTS bots can already approach perfect play on 13x13 with enough time on the clock and enough RAM.

  • lguser at 2019-10-04

    lazyplayer, how do you know that mootwo is mcts, also i do not believe that its just a 'few' moves until game over, if you look at games on here among top players, hex13 has about 30 moves per game on average

  • lguser at 2019-10-04

    With this win: https://www.littlegolem.net/jsp/game/game.jsp?gid=2123808 Mootwo is now tied 4-4 in games against LeelaBot, it seems that the title of best bot is unclear. Also Mootwo seems to be playing some unorthodox opening moves in its games.

  • lazyplayer ★ at 2019-10-04

    Iguser, the games last more moves but the outcome is already quite clear early on… :)

    For example the game you've cited, already here it's clear black is probably winning. Mootwo plays very good, I wouldn't say unconventional, indeed what's convention here? There are no conventions, humans have their own conventions but they've been refuted by the bots.

  • lguser at 2019-10-04

    these are very exciting times for hex, its a shame we arent discussing on a half decent chatting platform… i tried to do a move towards discord but it failed to gain traction

  • lguser at 2019-10-05

    Mootwo now 5-4 against leela, https://www.littlegolem.net/jsp/game/game.jsp?gid=2124434 this one had mootwo k12, leela j4,  j8. d10

  • Arek Kulczycki at 2019-10-05

    Good to see them struggling against each other!

    Gives me hope to try seriously to avoid mistakes and one day I finally get a win :)

  • Tom Ace at 2019-10-05

    I'm enjoying reviewing these games.  Thanks to all who are maintaining the bots here.

  • lguser at 2019-10-05

    yes, i can't wait to see bots master 19x19 too

  • Bill LeBoeuf ★ at 2019-10-07

    It looks like Daniel has secured a win and with the swap disadvantage !

    I think this will be the first bot loss to a human…

  • lguser at 2019-10-07

    I also think that Daniel seems to be winning. This would not be a first, lazyplayer beat leela in this game: https://www.littlegolem.net/jsp/game/game.jsp?gid=1999024  almost a year ago and daniel beat gzero in this game: https://www.littlegolem.net/jsp/game/game.jsp?gid=2033239 10 months ago.

  • lguser at 2019-10-07

    interestingly, both of those wins were also from a swap disadvantage

  • gzero_bot at 2019-10-07

    Wow - congrats Daniel.  Very similar to #2122370 - that might be inevitable given the opening, and leela will be playing similar moves since same model.

  • Bill LeBoeuf ★ at 2019-10-07

    Do the bots not learn from their mistakes?

  • Arek Kulczycki at 2019-10-07

    I have also tried to copy this game #2122370 once, but I was arrogant and tried to improve.

    Anyway this game is the most likely weakness of leela and Daniel used it brilliantly! Till very late I was not sure who was winning because there were potential tricks, but Daniel solved it all, big congrats!

  • lazyplayer ★ at 2019-10-07

    Oh damn, game 2122370 is identical, haha, this seriously devalues Daniel's win… but anyway, it's a weakness of the bots, they're not randomized! :D

  • lazyplayer ★ at 2019-10-07

    I'm also tempted to copy these games in my own game vs leela, I'll see… everything is fair game here… :D

    https://www.littlegolem.net/jsp/game/game.jsp?gid=2123807&nmove=6

    https://www.littlegolem.net/jsp/game/game.jsp?gid=2124448&nmove=10

    IgUser, my win was just luck unless it's repeated… I hope it'll be repeated but I've to be really careful, hoping/dreaming is not enough to win.

  • lazyplayer ★ at 2019-10-07

    Bill, the bots learn from self-play not from the actual games they play. And they've limited capacity to learn (due to their limited “brain” size), then you've to change the neural net (basically, change “brain” and pick a larger one) and start over with training. And training is costly from a computational point of view. In the case of Lc0, the chess engine based on neural nets. they're using donated computing power to do the training of the network, and despite this it took many months.

  • lazyplayer ★ at 2019-10-07

    Bill, to prevent issues like this, that is, repeated games, they need to be randomized somewhat, basically, to play slightly weaker moves for sake of making the bot more unpredictable.

  • lguser at 2019-10-07

    I wouldn't be surprised if a bot was in the works that used high level computing power, maybe will show itself in the next computer hex tournament.

  • gzero_bot at 2019-10-07

    Well the cats out the bag now.  :D  gzero (and AFAICT leela) have been playing the best move from move 2 onwards.  We've kept the illusion of randomness by continually improving the models.

    There is some randomness with MCTS when using multiple threads & virtual loss, but it isn't significant.

    Still I am actually very surprised it has taken this long to see very similar games (especially between the games between gzero and leela).

  • lazyplayer ★ at 2019-10-07

    gzero, don't be ashamed, actually that's the principled thing to do, no randomness, but always change opinion on what's best! :)

  • lguser at 2019-10-07

    it looks like the 2019 hex computer championship was won by a bot named DeepEZO, https://icga.org/?page_id=2908

    i think that if this bot were to play on littlegolem, it would be the best, sadly, i failed to find it online

  • Arek Kulczycki at 2019-10-07

    Ouch, I thought Daniel copied another game!

    Indeed this #2122370 one is entirely identical, eh.

  • lguser at 2019-10-07

    whats the point of that, Daniel shows he cares more about rating that improving at hex

  • Arek Kulczycki at 2019-10-07

    Well, no, It's bot's fault it played the losing game. Daniel did the right thing.

    I just wished it was a real human triumph over bot :)

  • Daniel Sepczuk at 2019-10-08

    Iguser,

    I played the most games with bots and lost most points and you write that i care more about points? Just look at my rating chart. Please, don't be funny (unkind you are).

    See my previous games against bots, maybe you'll notice something (especially when i played black).

  • Andrey at 2019-10-08

    Meanwhile new hex bot game has started link to 2125477

    Even for me who is far from even average hex level these bots games are interesting. Want to give some support to gzero because of sympathy it plays very strong not only hex but also other games. gzero is a monster - give me the rules, give me time and I will show you how shit you are :P (or how you should play in moderate words :))  But can't bet on gzero with dark horse mootwo which already gave some new and surprising ideas at openings. Will be interesting.

    Also I wonder about ability of leela and mootwo to play other logic board games. Personally I am interested in connect6 but it also could be other games. As far as I know here at LG plays strong twixt bot.

  • lguser at 2019-10-08

    gzero seems to have gotten stronger lately, I remember gzero saying that he was going to retire from little golem, I am glad that wasn't the case

  • Arek Kulczycki at 2019-10-08

    Depending on how fast will the game be I will try myself in a live commentator role :)

    https://www.littlegolem.net/jsp/game/game.jsp?gid=2125477&nmove=7

    So far it seems gzero on white has quite an advantage. K4 should be played and mootwoo's stones are almost useless!

  • Arek Kulczycki at 2019-10-08

    As usual for the bots mootwo started with a 4-4 corner which has been proven strong despite being less versatile than 5-5.

    gzero immediately took favour of the strong connection on the right edge and expanded with j2. Tactically correct, but it's debatable if that is the optimal move. People have been playing this for long time with mixed results, but maybe it just takes a tactical master like bot to use properly.

    I don't know what is the logic behind the mootwo's top-right corner fight becaues I see absolutely no gain for black!

  • Arek Kulczycki at 2019-10-08

    K4 was played indeed and mootwo tries to take a free nozoki on j3.

    https://www.littlegolem.net/jsp/game/game.jsp?gid=2125477&nmove=9

    But I'm not sure if gzero will allow this j3 for free - there is time to play tenuki like f7 or just escape from the corner with i5.

  • Arek Kulczycki at 2019-10-08

    The bots seem to pause for a while so I paste some random lines:

    Likely white win:

    http://www.trmph.com/hex/board#13,k7k7d10j2l3k3k5k4j3i5j5i6j7h8i9j4d5d3c3d2c2e6

    Expected continuation (I bet on white):

    http://www.trmph.com/hex/board#13,k7k7d10j2l3k3k5k4j3k2d5d4e4c12b12c11c9

  • Arek Kulczycki at 2019-10-08

    Eventually gzero played passive 10.k2 which made black j3 a free stone. Looks suboptimal to me.

    Interestingly mootwo secured bottom with h7 which gives white just one possible option to connect through bottom like: http://www.trmph.com/hex/board#13,k7k7d10j2l3k3k5k4j3k2h7h6g6g7f7e9f8f10

    If white cannot make use of the latter than maybe it's even black's advantage.

  • gzero_bot at 2019-10-08

    thanks for the commentary -  h7 was a very unexpected move from gzero's point of view.  moo is scary to play against, at least gzero can somewhat guesses leela's next moves. :)

    Going to play moo and daniel at hex 11, and anyone else that joins the waiting room.  note the model is only 3 days old… it might resign on move 3 or something silly!

    @Andrey haha - thanks for the support and monster compliments!! ;)

    @lguser - i don't know much about DeepEZO, but ICGA entries have been barren the last few years.  It very well could be the strongest bot.  I have a wee hunch moo is a mohex revival which historically won ICGA hex for many years.  I am 90% sure it isn't a pure A0 bot.

  • Arek Kulczycki at 2019-10-08

    12.h3 is another move from gzero that I don't like. How is it going to threaten anything on the left now? I expected e4 or some tactical magic in center. Anything that would make a direct threat.

    Black d5 is an obvious one here.

    Now white need to threaten simultaneously on the left and to surround h7, probably there is no time for this before black secures solid connections. We can expect a joseki now anyway, most likely d4-e4

  • Arek Kulczycki at 2019-10-08

    D4-e4 on the board. The last option to somehow delay the painful decision in center is white c12.

    The most optimistic line for gzero would be http://www.trmph.com/hex/board#13,k7k7d10j2l3k3k5k4j3k2h7h3d5d4e4c12c10b12e11e10d11f12c13f6g5g6i5h5i3i2h4f4c6b6f3e6d7c7d6d8e7e8f7f8g7

    In truth though white need tactical miracles in center with the use of both i3 and g7…

  • lguser at 2019-10-08

    arek, you seem to be calling out gzero's moves left and right but I don't remember you beating him :D

  • lazyplayer ★ at 2019-10-08

    Iguser, you set up a monetary fund as a prize for first person to beat this or that bot and then we'll do our job at that.

  • Arek Kulczycki at 2019-10-08

    Oh if only this gives me the right for it then after the champ I will arrange some games :p

  • lazyplayer ★ at 2019-10-08

    By the way, I think both i2 and j3 must be wrong. k2 is also very suspicious. The game seems black now.

    Ironically, humans play even worse initial stones than bots, or we just blunder immediately after the first few stones…

  • lazyplayer ★ at 2019-10-08

    I mean both J2 and L3

  • Arek Kulczycki at 2019-10-08

    Yes if by wrong you mean slightly imperfect :)

    I think humans just blunder randomly at all times

  • HappyHippo at 2019-10-08

    Thanks for doing this commentary Arek, very helpful

  • Arek Kulczycki at 2019-10-08

    And hooray the miracle center attempt is f6!

    Doesn't it simply lose to g5???

    I bet black g5 and then white tenuki in desperation like e11…

  • Arek Kulczycki at 2019-10-08

    Meh L4-L2 exchange obviously does nothing.

    In the meantime Im looking for tactics for white but in principle it must fail. If gzero does something here it's of 100% bot'ish nature.

  • Arek Kulczycki at 2019-10-09

    They're still playing but it's over. It's fascinating that there are positions quite clear for humans but the bots seem to have hope in. Mostly it's the other way round :)

  • gzero_bot at 2019-10-09

    @arek, after 26)d6, isn't it still pretty complicated?  It is for me! :)

  • Arek Kulczycki at 2019-10-09

    Ah true, I misjudged it! I wonder why mootwo didnt play g5!

  • Andrey at 2019-10-09

    Continue guys, please. Not for all humans it is clear :) What about humans - could we try to find out which move should probably gzero change (in case further moves will prove white lost)?

  • gzero_bot at 2019-10-09

    @Andrey gzero thinks it has been winning all along… so I am really curious to see how this pans out. :p

  • rookDriver at 2019-10-09

    @gzero_bot

    mootwo is not link to mohex ; it is another project, with mainly research purposes.

  • gzero_bot at 2019-10-09

    Thanks @rookDriver.  Are you allowed to provide any more info?

  • Arek Kulczycki at 2019-10-09

    https://www.littlegolem.net/jsp/game/game.jsp?gid=2125477&nmove=13

    this is the moment I was almost certain that gzero gave away the win.

    IF 16.f6 is a winning move then I think it may be the only one in the position along with maybe d12 that would just delay f6. Like I said in the first place, only magic tactics in center could save it for white and, as usual with bots, the tactics were found.

    Gzero if that's possible give us some insights how the bot would plan for 17.g5!

  • Andrey at 2019-10-09

    So, as far as i see white is going to finish right side connection with the very first move k7. And the left side seems done.

  • Andrey at 2019-10-09

    Like this

  • gzero_bot at 2019-10-09

    I'll run 17.g5 after the game.  I think gzero is about to go off on a tangent!

  • gzero_bot at 2019-10-09

    not easy to get a variation without changing the code… maybe moo can answer the question (since it was its move).

  • lazyplayer ★ at 2019-10-09

    Maybe 13 f4? But yes good game, as usual we humans were wrong, we underestimated white stuff on top-right…

  • rookDriver at 2019-10-09

    @gzero_bot

    It is a zero learning based algorithm, but we are trying to modify and improve some stuff :-)

    And we really appreciate all players here (real or not :)), it is really helpful.

  • lguser at 2019-10-09

    lately i've been seeing the obtuse 4-4 opening losing alot…

  • gzero_bot at 2019-10-09

    @lazy - thanks!  did you see the (ongoing) #2125583 11x11 game?   Think that might be the first bot v bot at 11x11.

    @rookDriver Good job so far! :)  I hope to read about your modifications & improvements.  I see you are playing 19x19, I thought it would take some serious compute to train that size of board, at least with conventional zero techniques?

  • rookDriver at 2019-10-09

    @gzero : Thx ; for 19x19, we try a faster technique, but it is really a test.

  • Arek Kulczycki at 2019-10-09

    @gzero I thought you could set up any position and start from there. We could start a new game with me as black and I play 17.g5 as soon as we reach the same position :D

  • lguser at 2019-10-09

    there's alot of demand for a downloadable hex bot

  • lguser at 2019-10-09

    gzero beat mootwo in this 11x game https://www.littlegolem.net/jsp/game/game.jsp?gid=2125583

    I wonder if both of the bots are better than the top human level in 11x

  • lazyplayer ★ at 2019-10-09

    Iguser, nobody is currently playing at 11x11 with any level of seriousness, I and Arek have played it some serious (training) games several years ago.

  • Arek Kulczycki at 2019-10-10

    lguser, you can trust or not, but this 11x11 bot game above is close enough to perfect play. Problem is that black had no chances since a11 was swapped… We already know that a11 wins and how it wins - it's almost “solved” even on 13x13.

    11x11 is a playground for tricks but bots would just play on diagonals which is the strongest but is solved.

  • Arek Kulczycki at 2019-10-10

    In fact in the above game I dont know how white win vs this variation

    http://www.trmph.com/hex/board#11,a11a11d5d4e4e8h7h6i6i5j5

    But anyhow, not interested :p

  • gzero_bot at 2019-10-10

    @Arek here you go

    after 18) d8…  white thought its chance of winning was 89%.  after looks like a mess to me ;)

    this is the old way of playing out the game, not the new penny way.

  • gzero_bot at 2019-10-10

    fixed link

  • Arek Kulczycki at 2019-10-10

    Oh, white d8 regardless of black move! This is really coherent then.

    I get it know! White strongly secures the left side in a way that it can connect to right alternatively with H6 (directly) or G6 (around H7 towards the K7) depending of black's need for i3!

    This is the brilliant tactics that I well knew had to be found but I didnt believe it existed :)

    Thanks for showing this gzero!!!

  • Andrey at 2019-10-10

    @Iguser, there can be demand for a downloadable hex bot. But think what effect can this bring to the games between humans. It can significantly reduce the interest amoung humans who want to compete with other humans and not bots. And especially it is strange when humans compete with each other using even not their bots. There will be disbalance between somebody who has bot and somebody who on his own. Also it will reduce the time of our thinking. And why we are here? We want to think, develop and try our best. The bots compete in special tournoments - and this is correct.

    At the same time I like that here are some bots. But it is important that bots don't hide under humans. It is true that playing against stronger opponent - you are growing. I admire owners of strong bots, but for me it is preferable to have an opportunity to play against bots and to compete with humans when you know they have the same instrument as you - only our brains.

    Also want to add that there is difference between draw game and game when one player has 100% theoretically win. I think we will miss hex 13x13 when it will be solved. Yep, we can go to the 19x19, but we will miss I think.

  • Arek Kulczycki at 2019-10-10

    Now the question is how a human can find that stuff - the combination of f6 and d8. Given that I had all the clues for what to look for in this position

    http://www.trmph.com/hex/board#13,k7k7d10j2l3k3k5k4j3k2h7h3d5d4e4

    I should have found this. I cannot recall what else I was loooking at…

    Another point is if a human should ever risk going into such position :)

  • Arek Kulczycki at 2019-10-10

    @Andrey indeed there should be two different modes to choose from: 1) game of human+bot vs human+bot or 2) human vs human

    I will trust the opponent does not cheat but there should be the choice because now using external help is not prohibited

  • lguser at 2019-10-10

    hex is the only game on little golem and probably the only established abstract board game where an 11x11 board is considered too small. I believe that if hex had a long history of play such as Go, it's default board size would be 35x35 or something around there. I am also surprised that an innocuous looking first move like obtuse corner would be almost solved in 13x13.

  • lguser at 2019-10-10

    but i forgot about twixt, in twixt, some players think that the default 24x24 board is too small

  • Arek Kulczycki at 2019-10-10

    @lguser I think the modern baseline for board size should be where the bots cannot finish training within a finite timespan so that we're always ahead :)

    But seriously, I think Hex 19x19 tactically is incomprehensible for human brain so it should be big enough no never become fully explored.

    13x13 is enough in these times where there are at most hundreds of players in the world. The obtuse corner move turns out to be really strong but with plenty of room for tricks between humans it will stay perfextly playable.

  • HappyHippo at 2019-10-10

    13x13 is a good size but it's becoming a bit too tactical at the higher levels of play. I find 19x19 too large. I think 15x15 would be great, but unless (until?) they add it to LG I'll be stuck with 13x13 ;)

  • Bill LeBoeuf ★ at 2019-10-10

    I certainly agree with Arek that bots are not going to solve 19x19 anytime soon.  But then we are all amazed at how quickly they have come to dominate 13x13 :)

  • lazyplayer ★ at 2019-10-10

    Iguser, smaller boards are too easy but large boards are too hard. I think it tends to become as hard as Go as board size goes to infinity. The “units” of Go are the cells of the board (the score is approximately the area of the board that you occupy or you've surrounded) and for hex it tends to become the same because you really need to “use” all the board to win on large boards.

  • lguser at 2019-10-10

    actually deep learning style bots get better relative to humans as board sizes increase so no board size is immune to bot domination

  • lazyplayer ★ at 2019-10-10

    Iguser, as board size goes up, brute force matters less, so there should be more opportunity for humans. I think the same was true in Go, the 19x19 was harder to crack for computers than 9x9.

    Anyway, you're correct overall that humans can't compete anymore against well-equipped bot. But I think we can learn from bots and we can improve and use bots ourselves as assistant.

  • lazyplayer ★ at 2019-10-10

    Same for time controls, LG time control allows better play by humans, although we probably need supporting stuff like books to store our analysis before we forget everything.

  • lazyplayer ★ at 2019-10-10

    I think you overall miss the point, the humans still win with much less computational effort, so, they're still using the better algorithm. The problem of course is that it's not good enough to compensate for hardware difference. Anyway, larger board, larger time controls, larger training efforts, etc etc, everything helps humans more than it helps bots. :)

  • Tom Ace at 2019-10-11

    I'd enjoy watching someone who's good at 19x19 play against mootwo.  The bot responds quickly.

    I'm not good, I got trounced.

  • lazyplayer ★ at 2019-10-11

    Tom, currently Arek is probably the least bad , but we're all very bad on 19x19. :D

  • lazyplayer ★ at 2019-10-11

    And 1 k5 is absurd, surely there must be fair openings on the first column…

  • lazyplayer ★ at 2019-10-11

    Uh, I was confused, 1 e11 was played. Even more crazy.

  • David J Bush ★ at 2019-10-13

    If we reach the point where many have a perfect opening repertoire memorized, another possible way to keep Hex interesting would be a more complicated opening protocol such as 3-move equalization. One player puts two black stones and one white stone on the board, and the other chooses which side to play. Or a larger grid of course.

  • lguser at 2019-10-13

    that doesn't help the memorization problem, one could gain an advantage from studying out of a 3 move start and playing it every time

    A better way to deal with the memorization problem might be to change the board size. Although a lot of the joseki patterns might stay the same with an increment of one. I don't know enough about how hex changes with different board sizes to tell.

  • lazyplayer ★ at 2019-10-13

    Iguser, the swap can be generalized in a way that I've discovered several years ago. First player chooses a number N0 and places N0 stones of any color, second player chooses if he wants to play black or white or if he wants to choose another number N1 such that N1<N0 and to place N1 stones and then first player has to choose and the process repeats with N2, N3, etc etc.

    This should be the optimal swap rule, making the full use of the possibilities of the whole board.

  • lguser at 2019-10-13

    lazyplayer, I don't see how that addresses the memorization problem brought up by david bush

  • lazyplayer ★ at 2019-10-13

    And anyway, I think this issue the bots are having here, lack of randomization, it's a flaw of the bots not of the game.

    The game as played in practice includes a “psychological” element of guessing what the other player will do and acting unpredictably whenever possible.

  • lazyplayer ★ at 2019-10-13

    Iguser, it addresses it because with my swap rule the game will always start from an arbitrary position.

  • David J Bush ★ at 2019-10-14

    Lazy, just because a starting position could be called arbitrary, does not necessarily mean memorization would not be involved. Consider chess 960. There are already opening studies on the 480 distinct starts. 3 move EQ is kludgy, but your proposal is the mother of all kludges. What a terrible thing to do to a pure innocent game such as Hex.

    My initial point was that as AI solves pie rule Hex, a strong player with excellent memory might be able to beat another player with more talent but who hasn't memorized the opening as well. A larger grid would provide a temporary refuge. So would a more complicated opening protocol. Ultimately we are all doomed. Have a nice day.

  • lazyplayer ★ at 2019-10-14

    David, It's mother of all kludge because, depending on what you want, it can gives birth to different babies! You say chess960 has 480 distinct starts? In my game, there are ~2128 distinct starts for 9x9 and ~2191 for 11x11! And the branching factor remains almost as high for the next moves unless both players agree on a starting position and they start playing conventional hex from there. So you've plenty of choices, indeed, you've ALL the choices possible while ensuring termination! :)

  • lazyplayer ★ at 2019-10-14

    And moreover, my rule approximately preserves the symmetry between first player and second player. Just think about it for some time you'll see it's the best… :)

  • lazyplayer ★ at 2019-10-14

    David, another property is that, in case normal 1-move swap is ENOUGH to balance the game, then my rule will *automatically* convert itself into 1-move swap equalization. Basically it does automatically pick the N so that N-move swap equalization is fair.

  • ypercube at 2019-10-14

    lazy player, it is not clear what values you propose for these N0, N1, N2,… values.

    If they are high, it looks to me as the game is altered to a point that it looks as a different game.

  • lazyplayer ★ at 2019-10-14

    ypercube, the players choose these, basically, first player choose an arbitrary position, second player choose if he wants to play black or white or if he wants to pick another position. The constraint is that the new position will have to contain at least one stone stone less than previous position. Then the other player will again choose between black, white or picking a new pos.

  • lazyplayer ★ at 2019-10-14

    Guys, should I be a coward and copy mootwo game vs leela in the current champion or should I play something else? Let's put this to a vote… :D

  • tangent at 2019-10-14

    Any good bot playing Gomoku Pro?

    I have no opponent.

  • Arek Kulczycki at 2019-10-14

    I say it's smart to play mootwo's moves if they are objectively good. Otherwise you risk leela changes and you end up with bad position. It's not cowardly to win :) You can play as many other games vs leela for glory as you want outside of the champ.

  • rookDriver at 2019-10-20

    Hello everybody,

    mootwo is trying to understand hex19, if anyone is interested in playing it, do not hesitate to send an invitation to mootwo.

  • Arek Kulczycki at 2019-10-23

    Eh I maybe will never beat leela if I failed this time. From my point of view I did everything right and I started from theoretically winning position of 1.a2. Still lost. I don't know how mootwo could win games given his ridiculous approach :p

    @Leela please I will appreciate if you show a piece of report from the bot. I wonder if the initial moves were to blame, that is #8, #10 and #12 or did I mess up later?

  • lazyplayer ★ at 2019-10-23

    If the guys running bots want to do us a favor, please analyze this position. It's our last game me vs Arek. Who had the theoretical win?

  • lazyplayer ★ at 2019-10-23

    Personally I think it's white win (Arek was white) via 18 f3. There are lot of complications but black is just playing tricks.

  • leela_bot at 2019-10-23

    For Arek's game against leela ( 2124651), leela thought f7 at move 8 would have been better than e8, but not by much at all, something like 46% vs. 52%. leela was right at 50% after moves 8, 10 and 12 (51.7, 53.5 and 51.4 to be precise). It got happier after move 14 (up to 61%) and from there just slowly and steadily went up (no big blunders!). Well played Arek!

  • leela_bot at 2019-10-23

    @lazy, leela thinks your position is a white win, but not by playing 18 f3, but 18 e4 (around 65% win, so by no means a sure thing). After f3 leela thinks c4 has a 54% chance of winning (for black).

  • leela_bot at 2019-10-23

    Did anyone else notice how thoroughly mootwo humiliated leela: It won on both sides of K12 and G11!K12: 2124434 and 2121893 G11: 2123807 and 2124448

  • lazyplayer ★ at 2019-10-23

    Leela, we all thought you had mastered hex and instead… it's more complicated. This is the funny thing of hex. It seems always almost solved but it's never solved.

  • lazyplayer ★ at 2019-10-23

    Btw, as humans, we also had the feeling that 14 was wrong. At least I had it. It's a simple local/tactical blunder. He believed it was a free stone but it was not free.

  • Arek Kulczycki at 2019-10-24

    Thank you Leela for the answers!

    Ah, 8.f7 then… I considered it but I thought it would be more mess and the further from edge the more tactical it would get.

    I'm curious about that 14.c8 because I would say that it is just a consequence of previous moves. Before playing #12 I had to assume that c8 must be free, else I lose. Given this assumption there is a question of WHEN to play it, and I tell you the logic that I always use. The most likely such a move (nozoki) is free when the stone on the other side, i.e. e8 is the most active. The big worry for white is that it slowly loses activity of e8, hence I thought the best moment was then at #14 when board still empty.

    Another point is that without c8 I cannot immediately build on d5, so I basically need to play far like 14.h3 which again becomes a mess… We as humans need simplifications and we try to achieve them at smallest possible cost, but it seems we cannot afford this cost vs the bots. In my next match vs leela I will try to avoid simplifications and play the mess but then I will probably make a ton of blunders ;)

  • Arek Kulczycki at 2019-10-24

     @lazy, leela thinks your position is a white win, but not by playing 18 f3, but 18 e4

    Wow, now that is interesting. I only considered f3, e2 and at last stupidly e6.

    And indeed I was thinking c4 was the best black attempt vs f3, hence I got e6 temptations.

  • Arek Kulczycki at 2019-10-24

     Did anyone else notice how thoroughly mootwo humiliated leela

    I don't look much at bot games, but I took a look now. It's very surprising to see leela losing. Worst thing is that mootwo doesn't play so well at all :D

    I like how this position shows why 4-4 is wrong, mootwo exploits it perfectly. (Yeah, Im a big 5-5 proponent)

    https://www.littlegolem.net/jsp/game/game.jsp?gid=2124434&nmove=19

    This is funny how leela loses at move #5!

    https://www.littlegolem.net/jsp/game/game.jsp?gid=2121893&nmove=5

    After this mootwo just plays extremely simple and forcing until leela goes into total desperation :D

    https://www.littlegolem.net/jsp/game/game.jsp?gid=2121893&nmove=23

    Next is what lazyplayer tried to copy. Really weird game. I don't know what is going on, but leela played 4-4 going fast into losing position

    https://www.littlegolem.net/jsp/game/game.jsp?gid=2123807&nmove=8

    and again soon total desperation :D

    https://www.littlegolem.net/jsp/game/game.jsp?gid=2123807&nmove=26

    And last is the most balanced but absolutely tactical game. I have no idea what the bots are doing, just nonsense but somehow tactically valid

    https://www.littlegolem.net/jsp/game/game.jsp?gid=2124448&nmove=18

    After this mootwo is just having fun. I need someone to tell me please that f2 was not the only winning move and there were others clean ones… It may well be however that f2 is indeed correct in this position, we just never see it because it's closely tied to k7.

  • Tom Ace at 2019-10-24

    Thanks to Daniel Sepczuk for playing mootwo at 19x19.   I would've liked to see moo play better, but 19x19 is hard.  https://littlegolem.net/jsp/game/game.jsp?gid=2128326

  • mootwo at 2019-10-31

    https://littlegolem.net/jsp/game/game.jsp?gid=2129789 is a more interesting 19x19 game.

    Thanks Arek Kulczycki for this game!

    Note that the model has now been specifically trained on size 19, this was not the case earlier, so there is reason to expect improved play.

  • gzero_bot at 2019-10-31

    Congrats to moo.  Out of curiosity I started training gzero for 19x19 also.  It has just began to learn that connecting both sides is a good thing!

    I'd be interested if it plays above 1500 elo right now, if anyone would like to challenge it.  It isn't ready to take on Arek&co. :)

  • lazyplayer ★ at 2019-10-31

    mootwo, 9 s15? :D

    this is nonsense but 19x19 is very hard for us to understand at all.

  • lazyplayer ★ at 2019-10-31

    mootwo, i rejected your game offer because i'm just not in mood to play hex and i already have too many games started. Another time.

  • lguser at 2019-10-31

    that was an unexpected win, i guess it goes to show that bigger board sizes do not increase the relative advantage of humans to bots

  • lazyplayer ★ at 2019-10-31

    Iguser, larger boards make “knowing” more important than “calculating”. In theory this should favor humans but in practice it'll favor computers because we humans don't have decades of time to dedicate on studying 19x19 hex.

  • Arek Kulczycki at 2019-11-01

    Honestly I made all my moves in seconds especially that mootwo was playing very badly in opening, but such is the game of Hex that much can be recovered. In tactics humans are no match for bots so it finally brute forced the win.

    Next time I will play more seriously but playing 19x19 seriously means a lot of work, might be over 50 moves

  • Force majeure at 2019-11-02

    Quick question - what do you think is the biggest bang for the buck, regarding hex skill development? I don't have much of a free time to study the game, however I would like to steadily improve. I don't understand bot's highly tactical play and somehow I don't consider it very useful for improving my game. Any ideas?

  • Arek Kulczycki at 2019-11-02

    Biggest is having faith in simplicity. However complex the position very often the best or at least decent move is the simplest one. Keep playing on diagonals and hold on to your stones and groups when attacked (don't let opponent undermine your stones by surrounding them)

  • Force majeure at 2019-11-02

    My question was more about how to learn in a smart way (not to spend too much time), but anyway thank for every suggestion :)

  • Arek Kulczycki at 2019-11-02

    Then I think my answer was the answer to your question, learn to play well by playing well, so you save time which you would otherwise lost on playing badly :D

  • lazyplayer ★ at 2019-11-02

    Force, copy the first few stones from the games of strong players (leela, Arek, Daniel, me, shalev), either by playing identical openings or nearly identical, and then as the game goes on, be more and more careful on what you play. Almost all games are decided near the very end, except at the absolute top level.

  • lazyplayer ★ at 2019-11-02

    Basically, don't get into the mindset, the game is already over, the other player know some tricks that I don't know and he has figured it out… it's almost never true.

  • lazyplayer ★ at 2019-11-02

    Force, I'm looking at your last game with leela, this one. I think you lost at 8, you had to play 8 L3. I agree with Arek, always keep it simple and logical. Maybe leela can tell us if 3 L3 was the right way to go or not. Anyway, well, with the bot you can't recover from a bad opening like that. That's why they've 99% win rate against humans. :D

  • lazyplayer ★ at 2019-11-02

    Another advice that seems to be valid is to compare relative worth of potential stones rather than trying to determine exactly who has the theoretical win. This should cut out some useless work. You'll quickly see that the simplest choices are that that go better with the other stones already on the board and with the edges.

  • lazyplayer ★ at 2019-11-02

    *are those that go better

  • lazyplayer ★ at 2019-11-03

    Leela, so what I did wrong in game #2124652 ?  I think my error was copying mootwo's bad opening!

  • lazyplayer ★ at 2019-11-03

    Well, more precisely, either the bad opening or the initial G7. Maybe I should have played J4 or I5 instead of G7?

  • Force majeure at 2019-12-12

    I'm afraid, that in the very near future the championship and rating list will be so dominated by bots…3 strongest players now are bots, I think there is 4th one that is climbing up rating ladder. With an ease to replicate them and possibility to cheat having access to such tool, it may no longer be such fun as it used to be :/

  • gzero_bot at 2019-12-12

    Who's the fourth?  gurgeh is me the human, and I suck - but I enjoy playing.  Now if I do win a game against a better player, I'll automatically thought of cheating. :p  And I have only cheated once, and it was more of an accident with #2137442, and it just made me feel really bad for doing so.  Daniel still managed to beat me, even with a couple of cheat moves. :)

    well gzero has retired from every game but 19, and I'll probably retire it soon also.  I was a bit miffed moo vanished as I was just giving it some friendly competition.

    but I agree, there was some initial excitement while leela/gzero coming through the ranks for the first time - but now it feels pointless, just as much as anyone can run a bot or cheat in the games such as reversi or chess, where super human programs existed for decades.

  • Force majeure at 2019-12-12

    gzero, you've misunderstood me - I do not accuse anyone of cheating. However, if there would be strong bots readily available to anyone, for many people it would be tempting and there is nothing to do about that. I don't remember the name, but in 3rd championship league there is a player with 8 games, all won including 2000+ player - I do not think that it is possible to progress that fast. However I have not seen its games, it may be also old player with the new nick.

  • lguser at 2019-12-12

    I dont think most of us want any bot to retire, some of us want the bots to stay out of tournaments but I think we are mostly fine with bots in rated games otherwise, please dont retire bots, they provide us with a valuable source of insight into a higher level of play

  • gzero_bot at 2019-12-13

    @lguser, FWIW I mean retired from further training and entering tournaments.  gzero will still be available for friendly games (accepting invites), or I can add to waiting room if anyone asks.

    .

  • gzero_bot at 2019-12-13

    lguser, FWIW I mean retired from further training and entering tournaments.  gzero will still be available for friendly games (accepting invites), or I can add to waiting room if anyone asks. [edited: formatting]

  • ypercube at 2019-12-13

    And lguser probably meant: there are many users who do NOT want the bots to retire from tournaments. Why leave them only in friendly games? Maybe(maybe) retire them from championship. But not other tournaments.

  • Arek Kulczycki at 2019-12-13

    I think there is still enough space for human players in tournaments. Recently, let's admit - ones of the most active in out community were you the bot owners.

    Also I think there has been more interest about bot vs bot games then ever about human games in this forum.

  • Arek Kulczycki at 2020-04-23

    Bot battle continued by a human :)

    http://www.hexmaster.net/#/games/2156495

  • Force majeure at 2020-04-23

    Interesting analysis, thanks for posting!

  • HappyHippo at 2020-04-24

    Nice work Arek, always good to read your insights

  • Arek Kulczycki at 2020-04-25

    I planned to share more games, but I was stupid. Only just now I have realized that gzero and leela truely play almost perfect on 13x13. I'm often able to choose 5 or 6 moves perfectly but not 20 to win an entire game.

    Good news is that those “strong” openings that are evaluated to be ~55% chance winning, like k4, are absolutely playable. It takes just 1 tiny innacuracy to equalize.

  • Arek Kulczycki at 2020-04-25

    *I meant m4 on 13x13

  • lguser at 2020-04-25

    I wish the bots were still playing against each other…

  • lazyplayer ★ at 2020-04-25

    Iguser, why they're no longer playing the infinity tournament? Eventually they would converge on top and play almost exclusively vs each other.

Return to forum

Reply to this topic