Collecting Data on Player / A.I. cascades

Unknown

I conducted an experiment for 1000 games played. ALL of these games were played with LThor/Hood/ProfX for consistency.


I marked down every time I got a cascade and every time the AI got a cascade. For this experiment I counted a cascade as 4 or more matches in one move.

A.I.
1839

Me
1221

Now I also recorded every game where I used at least one Health pack. 486 games

A.I.
744

Me
821

So in the 514 games where I did NOT use a health pack

A.I.
1095

Me
400

I realize this is a small sampling size relative to the total number of games played, but those are the numbers. I am going to start again tonight and record another 1000 games and see if the numbers move more towards even or if the stretch even more in the A.I.'s favor.

Unknown

I'd be curious to see how many cascades happened right before your victory by the AI

Unknown

mango_sentinel wrote:

I'd be curious to see how many cascades happened right before your victory by the AI

Noted, Will add it to the list

Jasonzakibe

Surprised there hasn't been more responses on this. It's certainly interesting and more data than most people seem to bring to the table.

atomzed

It is interesting definitely.

Can I check whether you take a match 4 whenever it occurs? Or did you prioritise certain colors over the match 4? That would make a difference.

I come to realize that the match 4 are really good in giving cascades and the AI *never* miss a match 4.

Buret0

atomzed wrote:

It is interesting definitely.

Can I check whether you take a match 4 whenever it occurs? Or did you prioritise certain colors over the match 4? That would make a difference.

I come to realize that the match 4 are really good in giving cascades and the AI *never* miss a match 4.

Yes, I have come to accept that when offering up free match 4s every turn to the AI when I'm playing with a maxed out mischief Loki.

Played right, you are essentially giving the AI 8 AP per turn, but then stealing 8 AP per turn. However, the danger is that the AI will come out ahead if they hit a cascade.

The data is interesting, but it may also be indicative of the playstyle of the player. When I'm not trying to force a match 4 onto my opponent every turn with a Loki/Cage combination, I tend to play very conservatively and I'm always very careful about what I'm leaving my opponent. I tend to split up grouped colors of tiles where possible (while still collecting a top 2 or 3 AP for my characters if available) to leave the AI with fewer possible chains and to make it harder to match colors I'm trying to deny.

That could be why I don't experience very many cascades, but experience righteous rage whenever I do get one.

Wonko33

Wasn't there a dude a while back that had come to the conclusion that you get more cascades too when you play like the AI?

Stuff like the AI always picks a match4 but we don't because we value 3 of a useful color more than 4 of another. And the AI favored vertical matches over horizontal ones or something? And it does matches that will generate another match 3 while we don't always because we are playing fast or we are just looking for certain colors.

I think that would make a difference too

Cypr3ss

I'd also be curious to see how often you get a cascade (massive cascade!) on your final turn, i.e. the one that ends the match. I've noticed I seem to get a *lot* of these, and end up waiting for the cascade to complete so I can get the 'victory' message and move on. It feels to me that my cascades and the AIs are similar in number, but the AIs are far more useful.

Regards,
Cypr3ss.

Eddiemon

Wonko33 wrote:

Wasn't there a dude a while back that had come to the conclusion that you get more cascades too when you play like the AI?

Stuff like the AI always picks a match4 but we don't because we value 3 of a useful color more than 4 of another. And the AI favored vertical matches over horizontal ones or something? And it does matches that will generate another match 3 while we don't always because we are playing fast or we are just looking for certain colors.

I think that would make a difference too

Also how are we scoring match 5s in L and T shapes? If I make one of them in a turn I will often end up with 3-4 matches before the AI gets another turn, betwene match 5, crit tile automatch and then another match 3 on my extra turn.The AI just makes a 4 in a row out of it and has more chance of a recorded cascade.

simonsez

Wonko33 wrote:

the AI always picks a match4 but we don't because we value 3 of a useful color more than 4 of another.

Unless you're trying to set up the AI's Loki, I don't think you're supposed to be bypassing match4s, unless it's an unusual case where the match3 is going to give you enough AP to hit a nuke.

Unknown

atomzed wrote:

It is interesting definitely.

Can I check whether you take a match 4 whenever it occurs? Or did you prioritise certain colors over the match 4? That would make a difference.

I come to realize that the match 4 are really good in giving cascades and the AI *never* miss a match 4.

With a non Loki team I prioritize Match 4 just like the AI because the odds of more matches after is higher.

aesthetocyst wrote:

Just to make sure I'm catching your implication...you're suggesting that if you're don't burn a health pack prior to a match, the computer is shifting the odds in its favor, dropping tiles in a manner more likely to cascade, in an effort to further damage your damaged toons, in an attempt to force you to burn a health pack ... or wipe out completely?

That is what the numbers show, yes. I am just posting the numbers that I got from my test. I am doing this test completely unbiased. I just want to show the results of the test. One of the things I was asked to track was how many cascades happened with and without Health packs.

Cypr3ss wrote:

I'd also be curious to see how often you get a cascade (massive cascade!) on your final turn, i.e. the one that ends the match. I've noticed I seem to get a *lot* of these, and end up waiting for the cascade to complete so I can get the 'victory' message and move on. It feels to me that my cascades and the AIs are similar in number, but the AIs are far more useful.

Regards,
Cypr3ss.

I was attempting to track that data, but it doesn't happen anymore unless your on a wave event. Now if you are facing a normal opponent in pve or pvp, the game ends when you down the final opponent. It no longer continues the matches on the screen. You will notice that as the bonus iso screen/points adding screen pop up there is a flash in the background of any remaining matches that didn't fire but they are happening somewhere else.

Eddimon wrote:

Also how are we scoring match 5s in L and T shapes? If I make one of them in a turn I will often end up with 3-4 matches before the AI gets another turn, betwene match 5, crit tile automatch and then another match 3 on my extra turn.The AI just makes a 4 in a row out of it and has more chance of a recorded cascade.

I only register a cascade as being 4 or more total matches. it doesn't matter if they are match-3, match-4 or match-15 tiles, as long as there is at least 4 matches in one cascade (if you match-5 and get "another turn" when the cascades stop that is a new turn")

Please feel free to request any other data you would like. I am going to start my 1000 matches again tonight around 6pm CST. All reasonable requests entered by that time will be added to the list of things to track.

Unknown

Thanks for doing this OP! I've been saying this for ages, but everyone just accused me of confirmation bias. Thanks for taking the time to record this, and very interested to see your results from the second go pf this.

As others have said, if you're already doing this and it's not too much to ask, try to record how many last call cascades the A.I. gets at the end of matches and after you've built up a power (if you have time).

Also, you may want to edit the thread title to something more descriptive about its content - for instance "Collecting Data on Health Pack Match Influence"

Unknown

The reason you won't get much response on this is that unless you play like an AI, the results are just too variable.

One of the problems with your team is that only Thor's Red and Yellow are going to create cascades. If it's not a mirror match, the results won't be exactly impartial. Even if it was a mirror match, it still shouldn't be equal, (eg you'd try to fire your yellow just after the enemy Thor does.)

Can you list your tile matching priority? Assuming something like 5's 4's
Are you farming the green before you use Thor's yellow? Are you purposefully matching to create cascades when you see it? Are you trying to get the AI to set you up for Xavier's Master Plan?

If you have this scenario at the top:



do you take the yellow match to try for the critical?

Do you find yourself favoring vertical or horizontal matches? (each has different probabilities of cascade!)


Most people are willing to accept that each tile is random. Will has specifically said this.

Unknown

theHappyDance wrote:

Thanks for doing this OP! I've been saying this for ages, but everyone just accused me of confirmation bias. Thanks for taking the time to record this, and very interested to see your results from the second go pf this.

As others have said, if you're already doing this and it's not too much to ask, try to record how many last call cascades the A.I. gets at the end of matches and after you've built up a power (if you have time).

Also, you may want to edit the thread title to something more descriptive about its content - for instance "Collecting Data on Health Pack Match Influence"

The Health Pack match influence was actually ancillary to what the goal was. I am doing this mainly to record cascade numbers. The other stuff is just what people asked for while I was doing it.

Unknown

traedoril wrote:

The Health Pack match influence was actually ancillary to what the goal was. I am doing this mainly to record cascade numbers. The other stuff is just what people asked for while I was doing it.

Ok, so "Collecting Data on Player / A.I. cascades" perhaps?

The current thread title isn't very indicative of the nature of what the post is about, which means many people might not look at it.

Unknown

daibar wrote:

The reason you won't get much response on this is that unless you play like an AI, the results are just too variable.

One of the problems with your team is that only Thor's Red and Yellow are going to create cascades. If it's not a mirror match, the results won't be exactly impartial. Even if it was a mirror match, it still shouldn't be equal, (eg you'd try to fire your yellow just after the enemy Thor does.)

Can you list your tile matching priority? Assuming something like 5's 4's
Are you farming the green before you use Thor's yellow? Are you purposefully matching to create cascades when you see it? Are you trying to get the AI to set you up for Xavier's Master Plan?

If you have this scenario at the top:
:yellow


do you take the yellow match to try for the critical?

Do you find yourself favoring vertical or horizontal matches? (each has different probabilities of cascade!)


Most people are willing to accept that each tile is random. Will has specifically said this.

Matching priority for me is always: 5's 4's anything that makes more than one match

I always go for the yellow in the scenario above, I am willing to let luck decide if I get the 5 match or the AI gets it.

I actually don't favor vertical over horizontal. I focus on what the visible result of my match will give the AI on its next move (Obviously I can't account for the random new tiles). basically I wont set them up for a visible power match or if say Cyc has enough yellow to cast on the next turn I won't line up teamUp tiles near red if I can help it.

When I said "anything that makes more than one match:






Even though I prioritize green over the other blue and teamup I will move the teamup down hoping to trigger more than just the two 3 matches some where in the RNG.

I realize that the results will vary player to player, but I can only do the test as myself and I wanted to share that with the community. I may play terribly and wouldn't know it, because I can only compare myself to the AI.

Unknown

aesthetocyst wrote:

I agree, he hasn't controlled for enough variables. To test his idea, that use of health packs influences the RNG of the following match, or any other similar idea siggesting the RNG is rigged against the player, he would have to player a sizeable number of mirror matches vs. the AI, with characters of the same setup, and also play like the AI, arbitrarily following a set strategy.

Some characters are more efficient at producing cascades. Following an objective, arbitrary, rule-based strategy as the presumably does, will produce consistent results over time. Assuming the RNG is neutral of course.

The OP needs to play the exact same match over and over, only altering the variable he wants to study. Prologue can be played indefinitely!

It is commendable that he played and tracked 1000 matches. That's quite an effort!

I disagree with this. My experience is the game adjusts and kicks your butt if you try to play the same match more than once or twice (my experience both in prolog and in events), but if you keep moving on between nodes it considers that playing as normal. You won't get accurate data from playing the same match over and over, unless you also record for playing different matches as well as a control.

I will agree there are possibly a number of variables that influence this so OP's data shouldn't be considered hard proof, but it's a very good start. My sense is that subsequent matches somewhat also adjust to your play-style. Use board shakers and it'll start giving you dry boards. Use tile generators too much, and it'll start giving crazy volatile boards that repeatedly wipe out all your tiles in a single cascade. There are other variables I suspect affect this as well, but those are the ones that most clearly seem to influence, based on my experience.

But it does seem that it wants to enforce taking damage (I suppose what it considers providing challenge), which would be fine in a classic PQ game where health packs were not an issue. In classic PQ games, you were rewarded for trying out new abilities etc.. since the only thing that happened was that you lost and could immediately play again. With the introduction of health packs, and apparent balancing around them, it becomes somewhat more frustrating for those of us who'd prefer consistent fair challenge, and not being pressured to pay to keep playing.

Unknown

I think a more accurate statement would be we don't know what rules govern the A.I., difficulty, and how cascades / boards are generated.

We can infer some things - such as character character levels weigh on tile generation (as well as strong colors), with similar levels favoring more stable boards while level differentials favor more volatile boards. The number of non-downed characters also have an influence (haven't counted the number of times where I've been completely wiped, or completely wiped the A.I. in cascades immediately after losing one character, but it seems to be consistent. Sometimes you're able to power through and counteract this by playing the odds, usually coming out of the match with the other characters at very low health, but sometimes not).

There seem to be a fair number of variables involved, but I also typically have fairly good intuition when it comes to this kind of stuff.

Unknown

traedoril wrote:

daibar wrote:

If you have this scenario at the top:



do you take the yellow match to try for the critical?

I always go for the yellow in the scenario above, I am willing to let luck decide if I get the 5 match or the AI gets it.

You're aware that this gives the AI the advantage in most cases? 14.7% your crit, 23.1% AI crit.

It's easier to play 100% probability moves, but do you also play to other probabilities? Eg:





where the right most vertical blue works best for cascades.

--edit updated probabilities.

Wonko33

simonsez wrote:

Wonko33 wrote:

the AI always picks a match4 but we don't because we value 3 of a useful color more than 4 of another.

Unless you're trying to set up the AI's Loki, I don't think you're supposed to be bypassing match4s, unless it's an unusual case where the match3 is going to give you enough AP to hit a nuke.

Yes, but it depends on a lot of things: what's on the line?, am I 3 short from a devastating attack and the match is there? , etc

The AI never passes a 4 match even if it meant he would get enough of an AP to crush you next turn, so it is a bit skewed

Wonko33

aesthetocyst wrote:

Cypr3ss wrote:

I'd also be curious to see how often you get a cascade (massive cascade!) on your final turn, i.e. the one that ends the match. I've noticed I seem to get a *lot* of these, and end up waiting for the cascade to complete so I can get the 'victory' message and move on. It feels to me that my cascades and the AIs are similar in number, but the AIs are far more useful.

Regards,
Cypr3ss.

That is a classic frustration. Kinda fun if you look at it as a victory celebration, but infuriating if you are shield hopping and in a hurry

Add the daily reward coming in and 2-3 gifts from facebook friends and you are mashing the screen screaming shield shield shield !!