Noticing obvious end game unnatural AI matches. Does AI count 4-5 matches and try to keep them even?

1

Enjoying the game, was a big fan of PuzzleQuest and this is a nice time waster. I even spent $5 for the daily gems and $5 for the daily souls to do my part.

THEORY: Each time the player gets a 4-5 match, a player +MATCHES variable accumulates. Using a color swapping team, the +MATCHES value can get +10 or +20 fairly quickly, once the AI only has 1 or 2 team members left, this +MATCHES value is used to increase/decrease the odds. If the player has a +MATCHES > 10, and the AI has < 3 team members left, this greatly (to the point of ridiculousness) increases the AI chance to have “random” drops match 4-5 and decreases the player chance has to have matches.

Not the end of the world, but my major annoyance is when I use a color swapping / soul farming team that towards the end of the match, the “random” drops become obviously manipulated towards the AI matching 4-5 of the color they need, and drops providing almost 0 4-5 matches for myself.

If you want to see this behavior yourself, load up a color-swapping team (Valkyrie / Alchemist / Dark Maiden / Naga Queen etc) and count the mid to end game matching behaviors. Game starts out ok with believable matches for each side, but once you have killed off 2 of the AI team, and have color matched some long turns of 4-5 matches, you will see:

End Player Turn with a match 3
AI match 4, AI match 4, etc, end turn with a Mana Attack or Skull Hit
Player has only match 3 options, or a match 4-5 of unneeded color. so match 3
AI match 4, AI match 5, end turn wiht a Mana Attack or Skull Hit

The last 10 matches I counted the end game matches and averaged results were:

Mid to end game: last 20 turns

Player: 2/20 “random” drops leads to a match 4 color that leads to a mana full and attack or another match 4 color
AI: 16/20 “random” drops leads to a match 4 color that leads to a mana full and attack or another match 4 color

The “random” drops become quite ridiculous with 15+ skulls scattered after an AI turn but none can be matched, or with a full board of tiles so that only one 1 possible match of an unneeded color is available.

Beating this is quite challenging and quite an accomplishment, but it becomes so obvious after the value of +MATCHES gets too high.

The drops should be random, they are clearly not and can be repeatably shown to not be random. Is this to discourage farming souls or just a an unintended result of trying to keep the matches “even and fair”?

Thoughts?

2

My thoughts? Random is random, and your organic pattern-matching hardware is kicking in and trying to put order to chaos. :slight_smile:

6 Likes
3

I’d agree wholeheartedly but the Console AI really does show some “catch-up” ala Mario Kart.

The majority of the time, if you open a battle with a lot of Mana gain, the AI will also get a similar cascade. The difference is the player gets a good starting board while the AI “catches up” purely on cascades.

It definitely makes you think the AI has other layers, I can attest. However, without access to the code, you have no proof and the development team has every reason to not divulge how advanced the AI is, so we will never know.

4

According to @Mr.Strange at the time, the discrepancy between the console AI and the PC’s was a bug. If that’s true, then you wouldn’t see anything nearly as complicated as the OP laid out above. As far as the PC goes, I know from previous experience that it’s quite easy to use rand() to make an AI that feels like it’s cheating against you, the programmer, so while it can be tempting to conspiracy-theorize, I am inclined to apply Occam’s razor to this sort of claim.

3 Likes
5

Yeah, I wasn’t agreeing with OP, because there are many easier ways to give the AI an “edge”, just relating how the Console AI randomness plays with your head since it’s broken.

1 Like
6

Just created an all goblin team which also starts off strong, lots of matches, and then as the match goes on, almost no match 4’s even though tons of green are being dropped by the spells.

Again not the end of the world, just makes for an interesting/challenging finale, but is clearly manipulated odds.

Boar Rider
Goblin Shaman
Goblin Rocket
Hobgoblin

7

You are free to believe what you like, but anecdotes are easily countered by other anecdotes. Just sayin’.

8

Hello trickycdr!

The good news is - you are totally right about the behavior. The bad news is - you’re totally wrong about the reason.

Gem matching/swapping games like GoW have some very well-established math about how large the board can be & how many colors you can support for a reasonable chance of matches. Too many colors and you won’t get matches most of the time. To small a board and the same thing happens.

8x8 board with 7 colors is in a wonderful spot. As all of you have experienced when playing GoW! Occasional “rich” boards with lots of matches, very infrequent “stall” boards with 1 or 0 matches possible. All part of the math.

Now here’s the fun part: If you can eliminate one color - via “remove all yellow gems” or “transform all brown gems to red” or somesuch - you’ve effectively changed the board math. Instead of 8x8 7 colors it is now 8x8 6 colors. Which makes the chance of accidental matches via drops & cascade matches MUCH higher. If you run a team with 2 such gem conversion troops, effectively turning the board into 5 colors, you can get some CRAZY chains going. (Random match chance on an 8x8 7 board is about 9%. Random match chance on an 8x8 6 board is about 15%. Random match chance on an 8x8 5 board is about 26%!)

But there’s no secret ai parameter adjusting the drop rate - YOU are adjusting the drop rate by removing colors, and thus making the board “richer.”

Of course, at the same time you make the board “worse” for random matches if the gems are of the color you have eliminated. And this is the key point: eliminating all gems of one color means new drops are 6/7 MORE likely to match/cascade, and 1/7 IMPOSSIBLE to match/cascade.

So if you make a very rich board, and then get the one gem color that fails to cascade, you’ve almost certainly set up the AI to do something ridiculous. No secret parameters or conspiracy necessary.

Here’s how you can mitigate this : be mindful of the colors your opponent uses. If you have a gem transformer, actively REMOVE the colors they use. That won’t reduce the chance that they will get lucky - but you can reduce or eliminate the chance that their lucky matches will be useful.

3 Likes
9

Are those percentages (9, 15, 26) assuming 7, 6, and 5 equally distributed colours? Not accounting for 1 or 2 essentially being doubled?

10

Those are rough estimates for random distributions (standard distribution) of gems across the board. So they don’t assume an even distribution - but they are averages for a variety of board states. Obviously each individual board would have a unique cascade chance, but those numbers give you a clear idea of how reducing colors broadly impacts your chances of getting a lucky match or cascade.

If you want to get into the math yourself, check out “map color problem” which serves as the basis for a lot of this thinking.

4 Likes
11

So what you’re telling me is it does not account for 1 or 2 of the types being doubled :innocent:

12

No - I was telling you that it does.

13

Well, doubling the amount of a colour is a specific state but you declared those figures were estimates of random distributions.