Page 2 of 2 FirstFirst 12
Results 11 to 17 of 17
  1. #11
    pas
    Member
    pas is offline

    Posts
    63
    Now searches depth 3 in around 2s.
    Also evaluation is more similar to the real game now instead of just maximising number of clears.
    Added 2 params to the program, water level and search depth

  2. #12
    SF Pleb
    jak8222 is offline

    Posts
    141
    Nice work! Coming along well.

  3. #13
    pas
    Member
    pas is offline

    Posts
    63
    Thanks,
    Here's what it looks like in game. Note that the puzzle detector doesn't detect puffer fish at the moment so I manually click it.
    Also the puzzle detector may be a bit buggy as it was rushed just to see it in practice.

    also note that only the logic is on github.


    I think I will stop committing now since the large hurdle is over.
    Last edited by pas; 07-28-2017 at 11:44 PM.

  4. #14
    SF Pleb
    jak8222 is offline

    Posts
    141
    What are your future plans for optimisations?

  5. #15
    pas
    Member
    pas is offline

    Posts
    63
    Well I was thinking of static analysis of the board first to get the rows that have the most colours close together.
    This way I can reduce the number of moves searched per depth by like half so it should greatly increase performance.

    I was also thinking to have static evaluation instead of counting clear combos (at the last depth move)
    E.g. go through the board and see patterns such as potential bingos and give them high score (a guess) without going deep enough to actually clear them.
    This way you're basicslly looking deeper without looking deeper.

    This last step i mentioned would be great and would work since it is basically the same as putting a static analysis bot (which are very cheap performance) within a depth first search bot.
    It would decrease the speed however will return a much greater depth search (estimated).

    Oh yeah and multi threading because it is easy.

    Note that this is all theory and that I will probably not implement them since I only wanted to come this far as I dont actually play the game.


    edit : also to add to optimisations, bilge bots generally look at repeated combos which are the same thing a lot.
    e.g. swap 0,1 and then 1,11 is sometimes the same as 1, 11 and 0, 1. This can be solved by using zobrist hashing on move index if the score of the move was 0.
    if you get rid of these then you will go deeper

    However that is very hacky to write and will look like shit code.

    I also feel like a monte carlo approach would perform better with tweaked settings due to the breadth of this game (60 swaps possible per turn is very large compared to say TD where it's like 5 or something).
    Last edited by pas; 08-03-2017 at 08:13 AM.

  6. #16
    pas
    Member
    pas is offline

    Posts
    63
    Surprisingly, I'm still not banned yet.

    I might write a patching bot soon

  7. #17
    Junior Member
    slikky is offline

    Posts
    3
    Quote Originally Posted by pas View Post
    Surprisingly, I'm still not banned yet.

    I might write a patching bot soon
    Hi dude, do you still have the bot? I dont know how to try your bot, but I would like to. Im on windows, do you have an exe?

Page 2 of 2 FirstFirst 12

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •