Why did the AI need a nerf?

124

Comments

  • babar3355
    babar3355 Posts: 1,128 Chairperson of the Boards

    And just to confirm... I was playing planar bridge yesterday and he had a straight line match 5 in green (I think) and a landfall match 4 in white (I think) available.  Regardless I checked his mana gains after he took the match 4 and they were identical between the 2 colors.  So he chose to take a match 4 over a match 5 which got him 1 less mana.  There were no servos, tokens, supports, activate gems, or anything else even on the board, much less affected   There were also no connecting combos on either line and since they were both horizontal they had the same odds at the top of the board.

    I mean, they clearly prioritize match 4's over match 5's now right?  All else the same? Seems like an intentional nerf to me.


  • Gunmix25
    Gunmix25 Posts: 1,442 Chairperson of the Boards
    shteev said:
    Gunmix25 said:
    DumasAG said:
    shteev said:
    Ohboy said:
    Before I go to the trouble of digging. Please state for the record that you do not recall constantly complaining about the incompetence of the coding.

    I hereby state for the record that any earlier statement Ohboy pulls out of the forums will be something I stand by, AND ALSO will be able to fully justify in the context of this thread.

    Go!
    That is some next-level politician stuff there
    It certainly is. He never stated for the record that he did not recall constantly complaining about the incompetence of the coding and stated that he would stand by any earlier statements he has made. A standard, I will neither confirm nor deny, stance.
    "..., there really doesn't appear to be anywhere in this thread where I said that Hibernum aren't incompetent, does there?"

    //Removed Inappropriate Content -Brigby

    "Aren't?" Yes, I agree. No where in the forums is there a statement by you saying that the people of Hibernium 'are not' incompetent, so at least we have that statement. That leaves us your likely opposing opinion, because you are certainly not neutral.


    On a different note: @octal9 your explanation of poor decision making makes sense. I've seen PW connected to red go specifically for supports and skip over landfall opportunities. Whereas I have seen green PW avoid supports with the exception of say an Animist's Awakening activation gem, in which it will try to swap it every single time. Blue appears to ignore supports and attempts landfalls; black is the same. Thank you for giving us the low down.

  • fiirst
    fiirst Posts: 438 Mover and Shaker
    edited August 2017
    octal9 said:
    shteev said:

    Wait, don't all L-shaped match5's contain a landfall?
    Fair. It would be more accurate for me to state that the L shape contained a Match-4.

    @octal9
    @shteev
     no landfall in that move ..... but not sure whether it is the only available matching at that moment....

  • babar3355
    babar3355 Posts: 1,128 Chairperson of the Boards
    fiirst said:
    octal9 said:
    shteev said:

    Wait, don't all L-shaped match5's contain a landfall?
    Fair. It would be more accurate for me to state that the L shape contained a Match-4.

    @octal9
    @shteev
     no landfall in that move ..... but not sure whether it is the only available matching at that moment....

    Erm, couldn't you move the 2nd from top dot down?  Which happens to be what Greg does every time now.
  • fiirst
    fiirst Posts: 438 Mover and Shaker
    babar3355 said:
    fiirst said:
    octal9 said:
    shteev said:

    Wait, don't all L-shaped match5's contain a landfall?
    Fair. It would be more accurate for me to state that the L shape contained a Match-4.

    @octal9
    @shteev
     no landfall in that move ..... but not sure whether it is the only available matching at that moment....

    Erm, couldn't you move the 2nd from top dot down?  Which happens to be what Greg does every time now.
    Oh! My bad, i did not ss log & screen that time....... so we can find it out
  • shteev
    shteev Posts: 2,031 Chairperson of the Boards
    I've always thought that maybe Greg didn't bother looking for match5s anyway, and any time it happened to hit on a match5, he'd just been looking for a match 4 and an extra gem just got taken along for the ride.

    Also, I've always suspected that perhaps there's a random element included in his moves which is why he sometimes does something weird that I don't expect.
  • Steeme
    Steeme Posts: 784 Critical Contributor
    I also noticed that the "hint" the AI gives you on your turn has also modified its behavior.  It told me ignore a five swap and take a lesser one.
  • Gunmix25
    Gunmix25 Posts: 1,442 Chairperson of the Boards
    Steeme said:
    I also noticed that the "hint" the AI gives you on your turn has also modified its behavior.  It told me ignore a five swap and take a lesser one.
    Saw this post and tested that out. He's right. I usually ignore that suggestion unless completely stumped. But yes, it recommends a lesser swap and frequently suggests swaps for loyalty gems.
  • shteev
    shteev Posts: 2,031 Chairperson of the Boards
    Gunmix25 said:
    Steeme said:
    I also noticed that the "hint" the AI gives you on your turn has also modified its behavior.  It told me ignore a five swap and take a lesser one.
    Saw this post and tested that out. He's right. I usually ignore that suggestion unless completely stumped. But yes, it recommends a lesser swap and frequently suggests swaps for loyalty gems.
    Do the hint and the AI use the same code to decide their moves? Or does the hint just pick any old move? @octal9?
  • octal9
    octal9 Posts: 593 Critical Contributor
    shteev said:

    Do the hint and the AI use the same code to decide their moves?
    Yep, but since the board state changes after your match, the AI will often take a different move
  • arNero
    arNero Posts: 358 Mover and Shaker
    Frankly, while I do understand what OP means about AI nerfing (quite a few times indeed I see AI miss out of match 5), given that they still get infinite cascades anyway, I don't see any real difference. Sure, they miss out match 5.... and then the next gems they get is literally match 5 anyway, so what nerf?
  • khurram
    khurram Posts: 1,090 Chairperson of the Boards

    [arNero] said:
    Frankly, while I do understand what OP means about AI nerfing (quite a few times indeed I see AI miss out of match 5), given that they still get infinite cascades anyway, I don't see any real difference. Sure, they miss out match 5.... and then the next gems they get is literally match 5 anyway, so what nerf?
    It is kind of avoiding (or not prioritizing) popping your supports. I had three of my supports that could be lined up but the AI did not even touch them for 3 turns until I just couldn't take it and blew them all myself. That's the first time that I have seen the AI not take such opportunity. 
  • tfg76
    tfg76 Posts: 258 Mover and Shaker
    When I started playing this game I was sure there was some bonus I didn't understand about match-4s. The AI seemed to be consistently going for them, regardless of other things. Now I generally try to lead the AI into match-4 traps where he's not matching his own colors, or not taking out my supports. Also, match-4s tend to lead to fewer cascades in my experience, so there's that too.
  • AresOmega
    AresOmega Posts: 75 Match Maker
    tfg76 said:
    When I started playing this game I was sure there was some bonus I didn't understand about match-4s. The AI seemed to be consistently going for them, regardless of other things.
    I would suspect that this may have something to do with many of the landfall effects found in the BFZ block. 
  • Steeme
    Steeme Posts: 784 Critical Contributor

    Hey guys, I've been paying attention to this a bit more, since I usually try and predict the AI's next move.

    I have a feeling the introduction to eternalize / aftermath is catching the AI's attention.  It looks like "Greg" is more interested in destroying activate gems than it is supports.

    When I have activates on board, they get removed pretty efficiently.  For example, when I pop Elspeth's second to generate two +4/+4 activates on board, the AI will attempt to remove them next turn, while at the same time ignoring my supports.

    When I have no activates on board, sometimes it will prioritize my support, sometimes it won't.  I haven't figured out what's up here.  However, I found it strange that it completely ignored a Deadlock Trap that I dropped until that support actually disabled one of its creatures.  It then proceeded to focus directly on removing that support.  That is, it looks like it changed priorities on the fly.

    This is quite anecdotal so I'm basically just sharing an experience.  Perhaps there is an internal "aggro" or "threat" meter which affects Greg's decisions.

  • Volrak
    Volrak Posts: 732 Critical Contributor
    I'm still yet to see convincing evidence for an AI having different modes.  Much simpler theories seem to suffice.  The vast majority, maybe all, of the AI's behaviour can now be explained by these priorities:
    1. Make a 4-match (exactly 4, no more), if present
    2. Match activation gems (this item might perhaps fit elsewhere in the list)
    3. Match its own colour gems, if present
    4. Match your colour gems, if present
    5. Make some other available match

    The AI sometimes hits your support or makes a 5 match, but when it does, it seems to be just an incidental result of following the priorities above.  Any definite observations to the contrary would be of interest.
  • babar3355
    babar3355 Posts: 1,128 Chairperson of the Boards

    @volrak Are you on slack or discord? I had a completely mind boggling one the other day where an opponent J1 ignored a straight line match 5 red and instead selected a match 3 led.  I could show you the board.. my screenshots always disappear after a day on the forums.

    No activates, no matches of either color available, 1 higher value match (green vs red) but it had his support on it.  He ended up taking the match 3 red at the top right of the board, so perhaps once they reach the "nothing better" point they don't differentiate between a match 3 and a match 5 thing and then just select the move at either the top left of the board or one of those two priorities?

  • Volrak
    Volrak Posts: 732 Critical Contributor
    babar3355 said:

    an opponent J1 ignored a straight line match 5 red and instead selected a match 3 led.

    No activates, no matches of either color available, 1 higher value match (green vs red) but it had his support on it.  He ended up taking the match 3 red at the top right of the board, so perhaps once they reach the "nothing better" point they don't differentiate between a match 3 and a match 5 thing and then just select the move at either the top left of the board or one of those two priorities?

    Yep - I think two possibilities for how a match is chosen if the "nothing better" step (step 5) is reached.  One is that all matches are considered as good as any other.  In that case, 5-matches might sometimes be picked at step 5 over 3-matches.  The other possibility is that 5-matches are still totally deprioritised; I.e. step 5 could be decomposed to 5.a "make some other available 3-match", and 5.b "make some other available match of any kind".
  • Gilesclone
    Gilesclone Posts: 735 Critical Contributor
    When in doubt they try to pick a match close to the top left corner.
  • Steeme
    Steeme Posts: 784 Critical Contributor
    Gilesclone said:
    When in doubt they try to pick a match close to the top left corner.


    I always thought it did this to "not disturb the board".  That is, if it chose something at the bottom, then it could potentially set the player up for a nice counter-move.

    I sometimes do the same thing by taking a harmless loyalty swap near the top.

Categories