Make Special Tile Placement Random (or less AI favoured)

Marvel Puzzle Quest MPQ Suggestions and Feedback
1

As it is right now any special tiles generated by players have a roughly 56% chance of being placed in a location that can be matched in one turn. Special tiles that are placed outside of a 1-turn match are far more likely to see matching tiles fall into the board in line with that tile. Decreasing the likelihood of a player special tile sticking around for more than a few turns.

AI special tiles on the other hand have a roughly 13% chance of being placed in a location that can be matched in one turn. Tiles that fall into the board have less chance of being matchable tiles to the special tile decreasing the likelihood of being able to match an AI special tile.

This does 2 things. First, it frankly frustrates me that there’s clear favour to the AI getting cascades, 4+ matches, and favoured tile placement for their special tiles. Why on top of all this advantage are you handicapping our roster too? All this does is increase the likliehood of the AI getting many AP without investing in any actual thought to developing the strategy of the AI to make it a more challenging opponent. It just becomes a frustrating opponent that I don’t want to play, because what’s the point when there’s significant, clear favour to the AI?

The second thing this does is that it devalues any character that does have special tile generating powers. Your deliberately handicapping 50% or more of any given player’s roster. Unless those power can generate many special tiles, or a cheap enough that it can be fired many times in close succession, what’s the point? Why use a character that generates a special tile when you know there’s a greater than 50% chance that any special tile generated is gone the very next turn. It’s far more favourable to use a character that just fires off the power without relying on tile placement to get value out the power?

If you’re going to give the AI all the advantages it already has for special tile placement, cascades, 4+, 5+ matches, please do not handicap us further with this preferred special tile placement. Just make it mostly random. It doesn’t even need the favoured placement the AI has. Just don’t deliberately place the tiles in matchable spots and increase the enjoyment I’m sure myself and other players would get out of being able to see some value in my roster characters that do generate special tiles.

If you won’t do that, at least lessen the AP cost of these powers, because there’s just really no value to using these characters.

2

75.9% of statistics are made up on the spot.

3

Citations?

4

Didn’t read anything more than title, but I don’t think you understand what random means.

5

I’m not sure exactly what you’re looking for here since I’ve been accused of making up stats. But I have in fact tracked my stats for 100 games after it seemed fairly obvious that there’s a clear favour of the AI to gain AP. The AI will almost always generate far more AP than you will. You can easily see this in your own games whether your tracking how they get the AP or not. This can only be achieved by making the frequency of events that generate AP skewed to the AI favour. These are the actual stats I have kept track of.

These are stats only against other teams where the AI is made up 100 of teams that can match tiles. for Match 4’s or greater I only tracked those matches where things just fell into place from off the board. I didn’t count a match 4 that fired because of 3* widows ability to place green tiles on the board, or where a match was made that lined up the next move to be a match 4+. This only tracked matches that magically came from off board.

If you don’t believe these stats. I’m happy to post a blank of the excel spreadsheet I used. We can crowd source this. 10 people playing 10 games should yield similar results. If you want to get more accurate the larger the sample size the better the results will be. This is just been my own tracking.

My Cascades: 260
AI Cascades: 385
My cascades/game: 2.6
AI cascades/game: 3.9

My Special Tiles created: 355
AI Special Tiles created: 120
My special tiles matched in 1move: 200
AI special tiles matched in 1move: 15

My natural match 4’s or greater: 100
AI natural match 4’s or greater: 135

6

You know, even if 100 games is a good sample size (which it probably isn’t), this behaviour could actually be caused by the ‘kinds’ of matches that you as the player choose to make (which are not strictly random) vs the kinds of matches that the AI makes (which is fixed by pretty limited AI behavior).

It might be interesting to see how your matches turned out if you always made the ‘suggested’ match every turn for 100 games (ie, the match that flashes if you wait a few seconds). What you should really be tracking, though, is simply the quantity and color of all the tiles that are generated to fall. We would expect each color to be equally represented, and that’s about the extent of it. The effects that those randomly generated tiles have on the board depend entirely on the non-random current state of the board.

You realize that it would actually take extra code to give the AI this kind of advantage, right? And even more code to give them the advantage while still making it plausibly seem random.

The only advantage the computer has is that they start every match with full health, every time.

7

There’s a 96.87% chance that I’m never going to look at this thread again.

8

Just and obligatory Gambler’s fallacy link for those who want anything useful in this thread Gambler's fallacy - Wikipedia.

Basically with your thinking you would assume that after 100 heroic tokens that resulted in 2* character you have better chance of having 3* from next pull. Randomness doesn’t care what the previous results are, every result could go any way, that’s the point of randomness.

9

It’s definitely not random. And while 100 data points is not a large sampleset, I don’t see anyone else taking me up on the offer to crowd source the data gathering. The larger the sample size the more true the results, but from my sample size there is a clear lean towards not only the AI acquiring significantly more AP, but also a more favourable placement of special tiles.

If you don’t believe me, just pick a character that generates special tiles. Iron Man, Punisher, Daken, etc… fire off the power and just see where the tiles go. If it’s possible to place the tile in a spot that is matchable in 1 move it will likely go there. Do this over 100 matches. They will definitely show a clear leaning towards being placed in a spot that is matchable in 1-move. When you play the same characters against the AI, make note of where the tiles go and how many of those are able to be matched in 1 move.

You can say I’m wrong all you want, but if you’re not looking for it, you’re not going to see it. You don’t want to believe my data, but you don’t want to be part of verifying the data… there’s no helping your point of view, but if you pay attention, and you want to part of verifying the data. I’m happy to see other real data other than my own. If I’m wrong I’m wrong, but I’ve actually put up the data. If you want me to be wrong, why don’t you take part and share your data?

10

That’s true… but the opposite of that is to admit to having a confirmation bias.

Also, for tile-creators like Daken, the placement of a tile in an immediate match situation depends totally on the current (non-random) state of the board. If there are only 4 red tiles on the board, and 3 of them make a match, that’s a 75% chance of getting a tile in a matchable spot. If we’re talking 3* Daken, you’re guaranteed to have one of your two tiles in a matchable spot.

11

I’m legitimately baffled by this post. I’ll assume you’re not trolling though. First of all, did you REALLY keep track of every single random tile generated for both players and the computer, then calculate how many ways that tile could be matched in one turn? I doubt it. But even if on the wild chance that you did, what about a two to three turn match? Four turn etc? You’re not considering all the possibilities would matter statistically making your conclusion invalid.

12

I’m not trolling. I played for quite a while and I felt what seemed to be the AI getting an odd number of match4+'s and I thought, well, randomness has strings of patterns in them, but it just kept going and the more it happened the more I noticed it. So I thought, I’m a rational personal. I can prove to myself that I’m just seeing a pattern where none exists because of a run of bad luck. So I brought up a simple spreadsheet and started tracking the match4+'s. I did that for around 40 or so matches and once you start tracking things you start seeing other behaviors. So I thought I’ll start tracking these other things too just see if I’m imagining things.

I tracked match 4’s for longer than my special tile matches because I didn’t start doing that until later on, but yes. I did really, in fact track these data points. 15 minutes here and there in the morning or afternoon, or on a weekend, etc… it adds up quick.

I did in fact keep track of every single special tile generated in a match where it was relevant. I didn’t track Storm’s one where she basically fills the board with attack tiles, but for characters that only generate 5 or less special tiles on a power fire (which is almost anyone) it’s not hard to track.
Step 1) The power fires; Did I fire the power or did the AI fire the power, tick a box
Step 2) how many tiles was generated, count, record in the next box
Step 3) how many of those tiles generated can be matched in exactly one move. Record
Repeat until you have a pile of data, divide the result by how many games you played, then you can see if you’re imagining things or maybe there’s something there.

This is not a difficult thing to track. When you think about powers that create countdown or other special tiles they only generate 1-4 or so. When it fires off you can easily see if that tile can be matched the very next move. Just record the info before you make the next move.

Same thing with tracking match 4+. It’s not hard to track if you limit yourself to relevant data. I only tracked the data where there was no possibility of a match 4+ until the tiles fell into the screen from a previous move. It’s not worthwhile to track match4’s that happen because of a move myself or the AI makes. That’s just a choice of match making. It is worth tracking if you make a move and the tiles that come on screen happen to perfectly line up for the AI’s move to be a match 4+, or if a match is made and it creates cascades of matches from tiles that fall onto the board from offscreen. Neither myself or the AI were aware of what the tiles coming onto the board are going to be. So if there’s a skew towards the AI tiles that fall onto the board should generate more cascades than for my turn. My data would seem to say, yes, there is something there. This happens on average few times per game, so it’s not like that’s a hard thing to track either.

All this data shows is that the placement of tiles that come from off screen are not random. They’re skewed to allow the AI an advantage in gaining AP. The placement of your own special tiles are also not random. They’re skewed to be matched in one move. The placement of the AI’s special tiles are also not random. They’re skewed to not be matched in one move.

Aside from asking yourself why would I track this data (because it shows the game is handicapped against you), you might also ask what would be the point of skewing the placement of tiles or AP gaining in favour of one or the other? The reason would be simple. It’s much easier to program an algorithm that knows if its the player or the AI’s turn and places the tile in a matchable or not matchable spot than to program and AI that knows and understands what each of the powers its playing can do and how they synergise or not with other team members. An algorithm to choose where to place a tile is much easier to program than to create an AI that can make intelligent decisions.

The most intelligent decision the AI currently makes is that if two or more characters have the same AP colour power that it will try to store up AP until it can fire the higher cost power. This is overruled if either the higher cost power owner is stunned or dies, or X-number of turns pass by (haven’t bothered counting the turns). then it just fires what power it can.

13

There are a ton of factors you’re not considering, some of which you can see others mentioning above. I’m going to focus on this particular incorrect assumption.

You are correct that the AI tends to prefer more expensive powers, but completely incorrect that there’s a turn counter that expires and tells the AI to fire the less expensive power.

On any given turn where the AI has enough of a single color AP to fire power A, but not enough to fire power B there is a random chance to fire the less expensive power. This random chance is influenced by the difference between the AP pool’s current size and the cost of the more expensive power.

If you had tested your assumption, you’d have noticed it was wrong, as the number of turns before firing the cheaper power would have varied. And that’s just one of the assumptions you were prepared to admit to.

14

so that minor picky part still proves the original point. The AI doesn’t know how to make smart decisions that’s why your tile placement is in fact skewed and the generation of new tiles as they come onto the screen after matches is skewed towards the AI generating more AP.

At first everyone’s saying prove your data. So I show my data. Now everyone’s saying the data is wrong, it’s not large enough sample set, it’s not gathered correctly. Well, let’s go with what I suggested. Let’s crowd source this. If I want to track how many special tiles I generate are able to be matched in 1 move vs. how many the AI generates. That would seem to be a straightforward process. Fire the power. Look at the screen, count how many tiles can be matched the very next move, record the information.

Same thing for match 4+ or cascades of matches. Only count the matches that happen because of new tiles falling into the game screen. No one has any knowledge of what those tiles are going to be. It could generate new matches (a cascade) or not. If there’s a skew towards new tile placement that favours the AI gaining AP, this will be another trivial count.

The time you have spent ignoring the message and data points, could have been spent gathering actual data trying to disprove my data. If you think my method is flawed, let’s get together and really discuss this. I don’t expect or want anyone to take my data at face value. The only way the game will change is if in fact you also gather data. The more data gathered, the more true the results. If this was crowd sourced you could have doubled or tripled my data points by now and you would actually have definitive proof. I’m either wrong or I’m right, but to go in immediately and ignore the data is not the way to tackle a potential problem. It wouldn’t surprise me if my percentages were off. 100 data points is not a lot of data, but it’s enough to show that there’s significance that’s probably worth a deeper look.

15

Finally got back to this and fair enough. Let say you’ve accurately tracked data and it is true that the AI has a slightly higher favor for random special tiles. We can assume this has been in for a very long time, and the computer AI is balanced around this factor. Is the AI currently crushing people? No, in fact if you play carefully, you’re designed to win battles with a variable amount of health attrition. In fact, the game designs the AI to not intentionally match match fives if they have the option of turning it into a match 4.

AI programming in general can be difficult and quite cpu intensive. Finding shortcuts like this that balance out as a whole is not a crime in a game made to be run on a phone. So if it is true about what you say, why is it to the health of the game that this be changed?

16

The data is interesting, but I can’t understand why it would be true.

The enemy AI constantly does dumb crap like use Human Torch’s Green and then immediately match it. If they were going to go through the trouble of putting tiles in good positions, you’d think they’d also go through the trouble of making sure the AI doesn’t match their own tiles. Also, there are times where I have countdown tiles that are easily matchable and the AI will just ignore them. Again, if they’re going to make it so my tiles appear in bad positions, why doesn’t the AI prioritize matching them? Even when the AI has abilities where they can choose where the tiles go (e.g. 3* Cap), they still use them poorly. If the developers had code that could identify good and bad positions on the board, why not use it for these abilities to make the AI play better?

A development team has limited time to add features, and I’d be confused why they would prioritize something like this. It makes the game somewhat harder, but so would making the AI smarter. I don’t see the advantage in writing cheating code as opposed to some really simple fixes to the AI behavior. These fixes would make the AI appear more human, would make certain characters much more useful on PvP defense, and would generally make the game more fun. Cheating needs to be unnoticed by the players, so all it does is make the game slightly harder. If they just wanted to make the game harder, there are several other ways they could do this that would be less effort.

So because of that, I have to assume there are flaws in the data. Possible ideas:

  • When you make matches, you prioritize certain colors. This means you might end up making matches near the top of the board which are less likely to cause cascades. Or you might pass up a match-4 since it’s in a bad color, but the AI will take the match-4 and these are more likely to cause cascades.

  • The AI will always take a match-4 if it is present. Perhaps you aren’t as good at noticing when they are available. The AI makes more of these, so they get more cascades.

  • The AI has abilities that encourage cascades. e.g. Ultron minions have passives that cause board shake up, so it could prime the board to be in a state where a cascade is more likely.

  • All abilities that generate special tiles are not created equal. e.g. Daken creates a red strike tile when green is matched, but Bullseye creates a purple protect when purple is matched. We would expect Bullseye’s tiles to be harder to match than Daken’s tiles, since to create them we need to remove the color that is used to match them. If you want to do these numbers, you need to make sure you’re using the same characters to avoid these possible biases.

17

Lots of good points here, but this one is particularly important. You can’t run these numbers as AI vs. human. You have to track it separately for every skill in question. And even then, the different match prioritization will influence available matches.

All the data in the world doesn’t matter if the methodology in gathering it doesn’t hold up.

18

Every now and then I debate whether or not I should just start recording all of my matches to show the shenanigans I see the AI pull. Then I think better of it knowing that the AI will probably know I’m recording and pull back on the shenanigans, making me to look like the fool.

…