Emrakul's Corruption: AI Cascades

Sorin81 · January 2017

Whoa.

Maybe I'm off my game, or maybe it's just a series of bad luck with this event, but I can't seem to gain any ground.
I'm a decent player and I feel like my decks are pretty darn good (not all mythic, platinum level good, but good). Typically I win more than I lose.
What I've run into with this event more so than before is the ai having the ability to string together insane amounts of mana cascades. Leaving me with very little to work with.
Has this been a problem for anyone else in this event?

Springs · January 2017

Agreed, I had the A.I play Oliva and then deploy the gatewatch turn one.. Three boomship on the board, 71 points damage.. seriously???

Ampmp11 · January 2017

This happened to me in terrors which was even worse with the enraged mechanic. It does seem like cascades are happening with much more frequency though ...

Sorin81 · January 2017

Ampmp11 wrote:

It does seem like cascades are happening with much more frequency though ...

I have noticed that too. Not just for the AI but for me as well, though it does seem to favor the AI more.

Steeme · January 2017

Played two matches and barely scratched by. Missed objectives in both. I just figured the matchmaker was trying to make me lose, not sure what else to think of it.

But yeah, in one turn, the AI had such a glorious chain-combo that I had completely written the match off. Was not impressed. Luckily, his deck wasn't very good (I mean it was just another deck that relies on Bacon, same old, same old). I had Frog and Moon already filled so I ended up surviving.

My point? Wasn't expecting it, and hadn't seen such a glorious combo in ages. Definitely out of place.

Stormbringer0 · February 2017

Same here. Not in every match, but in two games I just got wiped out.

The problem is not only the massive cascades, but the lack of combining your colors. If you can get no ground, no matter which PW you're piloting, you cannot win. On the other Hand, if you're cascading too, it's not to hard to outclass an heavily cascading AI with a good deck.

Volrak · February 2017

I'd say my games have been fairly normal cascade-wise. But that's completely non-notable and almost too uninteresting to post. Almost... except I also get the chance to explain to anyone who doesn't know what confirmation bias is: it's this thread.

buscemi · February 2017

Volrak wrote:

I'd say my games have been fairly normal cascade-wise. But that's completely non-notable and almost too uninteresting to post. Almost... except I also get the chance to explain to anyone who doesn't know what confirmation bias is: it's this thread.

Mersicide
Mersicide
Mersicide
Elvin1972
Mersicide

Sorin81 · February 2017

Volrak wrote:

I'd say my games have been fairly normal cascade-wise. But that's completely non-notable and almost too uninteresting to post. Almost... except I also get the chance to explain to anyone who doesn't know what confirmation bias is: it's this thread.

Confirmation bias? This thread? No freaking way!
Wait a minute...

Sorin81 wrote:

What I've run into with this event more so than before is the ai having the ability to string together insane amounts of mana cascades. Leaving me with very little to work with.
Has this been a problem for anyone else in this event?

Yeah...Maybe a little.
Hmm. The question does seem to beg for confirmation. Perhaps I should have expanded on it with something like
....or has this event been almost boring and non post-worthy?

Astralwind · February 2017

I take AI cascades as a consideration into my deckbuilding too.
Gotta have lots of creature removal or control. And gotta keep them coming in non-stop to stand a chance.

Volrak · February 2017

I should add - It's of course not impossible that the devs spent weeks working on a gem drop algorithm whose effect is to control how often cascades happen, in such a way as to produce a result that's statistically indistinguishable from what you'd expect if gem drops were simplistically random. But if that's what they really did, then based on the evidence so far, it was even less useful than making the UI blue.

I should also add that more evidence is always awesome and I don't want to discourage that aspect.

buscemi wrote:

Mersicide
Mersicide
Mersicide
Elvin1972
Mersicide

Wrong thread perhaps.. nobody's talking about matchmaking, which is known to have a strong non-random component.

Sorin81 · February 2017

Volrak wrote:

I should add - It's of course not impossible that the devs spent weeks working on a gem drop algorithm whose effect is to control how often cascades happen, in such a way as to produce a result that's statistically indistinguishable from what you'd expect if gem drops were simplistically random.

I guess that would explain what they've been working on, rather than working to fix other known issues.

madwren · February 2017

We are hardwired to remember negative experiences much more strongly than positive ones. Just human nature.

That being said, for awhile I was convinced that red's destructo-cascade effects seemed to work better for the AI than myself. You know, I drop a Volcanic Rambler or an Abbot, nothing. I trigger Koth's first ability, nothing. AI does it, and gets multiple mana matches each time, filling his cards and emptying his hand. I did once count up to 10 or 11 in a row where I got zero mana, whereas the AI never received zero mana, but in a small sample there are always runs like that, and for all I know I got it the next 11 times in a row.

So, while I'm 99.9% sure that the AI doesn't have an advantage, and while I'm aware that there's clearly a perceptual bias, it still FEELS that way every time I play Koth or Chandra, and it still irritates the hell out of me every time I trigger a cascade and get nothing for my effort.

On the counterside, if the AI had words, I bet it'd be swearing at the number of times I had a first-turn Deploy with Nahiri, while I've never run into a first-turn Deploy from the AI.

AngelForge · February 2017

I would suggest to get rid of the possibility to cascade, so it really is more a puzzle quest than a luck quest.

Ironwolf · February 2017

I definitely have noticed an increased rate of cascades and ranted about it during that event. I have an 87% win rate and lost 6 out of the initial 9 you start with due to the insane cascades. The AI was casting 2-3 cards per turn.

It is definitely programmed to do so. Most often the AI has ignored it's own color matches to get landfalls, but I've noticed it has been making color matches that seem off at first (e.g. red for Kiora) only to have it cascade 7 times. And I'm not talking the combinations you can see and plan to cascade - I'm talking about the ones that are off the board that the AI only knows.

I get that they have to program the AI a little to offset the human factor and not storing loyalty, not eliminating weak creatures on the board (from say, a Frog or Moon), but I think they balanced it a little too far in one direction that event.

Volrak · February 2017

Ironwolf wrote:

I definitely have noticed an increased rate of cascades and ranted about it during that event. I have an 87% win rate and lost 6 out of the initial 9 you start with due to the insane cascades. The AI was casting 2-3 cards per turn.

It is definitely programmed to do so. Most often the AI has ignored it's own color matches to get landfalls, but I've noticed it has been making color matches that seem off at first (e.g. red for Kiora) only to have it cascade 7 times. And I'm not talking the combinations you can see and plan to cascade - I'm talking about the ones that are off the board that the AI only knows.

I get that they have to program the AI a little to offset the human factor and not storing loyalty, not eliminating weak creatures on the board (from say, a Frog or Moon), but I think they balanced it a little too far in one direction that event.

I didn't see what you saw, but I find it hard to believe that the AI is programmed to cheat on gem drops. Partly because in the long run I see as many off-screen cascades for myself as for the AI, and partly because if the devs want to increase AI skill, it's easier (and less likely to generate player complaints) to program the AI to prioritise matches which will trigger on-screen cascades. They haven't done this. For example, the AI generally always chooses a non-cascading on-colour 3-match over an off-colour 3-match which would have cascaded into an on-screen, on-colour match.

When the AI picks an off-colour 3-match over a 3-match of their own colour, they're generally doing it to hit your support or a special gem of some kind (e.g. an activation gem). If anyone captures video where that isn't the case, I'd be interested to see it. As for 4-matches, the AI always seems to choose 4-matches of any colour over 3-matches of their own, even when it gives less mana. Setting up off-colour 4-matches for AI Koth can be a great tactic to trick him into ignoring all the red matches on the board.

buscemi · February 2017

Volrak wrote:

I should add - It's of course not impossible that the devs spent weeks working on a gem drop algorithm whose effect is to control how often cascades happen, in such a way as to produce a result that's statistically indistinguishable from what you'd expect if gem drops were simplistically random. But if that's what they really did, then based on the evidence so far, it was even less useful than making the UI blue.

I should also add that more evidence is always awesome and I don't want to discourage that aspect.

buscemi wrote:

Mersicide
Mersicide
Mersicide
Elvin1972
Mersicide

Wrong thread perhaps.. nobody's talking about matchmaking, which is known to have a strong non-random component.

See also this thread:
viewtopic.php?f=31&t=58559

I think it's highly relevant when talking about shonky RNGs in a game to consider other shonky RNGs in the same game.

We don't *know* that matchmaking has a string non-random component... we merely suspect it from the data gathered. D3 has made no comment on the subject.

Remember we're talking about the coding team that brought us this:
viewtopic.php?f=31&t=40244&p=492761#p492761
perhaps because true randomness is 'unfair'?

Volrak · February 2017

buscemi wrote:

Mersicide
Mersicide
Mersicide
Elvin1972
Mersicide

See also this thread:
viewtopic.php?f=31&t=58559

I think it's highly relevant when talking about shonky RNGs in a game to consider other shonky RNGs in the same game.

I'm not concerned about the possibility that matchmaking biases imply systemic biases in all random game events, because underlying RNG library calls are simple and well tested (by thousands of apps), while any event-specific code (e.g. matchmaking) by definition won't affect other types of RNG result (e.g. card drops). But you can certainly argue that buggy code in one area of the game implies the possibility of buggy code in a similar area (although I'd argue in turn that card drops and gem drops are much simpler than matchmaking, on the assumption that there's an actual intent to select opponents according to some additional non-random constraints). Of course, if the devs would explain the matchmaking algorithm then we'd know whether what we see for matchmaking is buggy or deliberate.

buscemi wrote:

We don't *know* that matchmaking has a string non-random component

What we do know is:

The chance of facing an actual random opponent from a group of 3,000 is 1/3,000
So, if unbiased, the chance of getting the same opponent even twice in a 30-game event (birthday problem!) should be only 14% - one in 7 events.
And the chance of facing the same opponent 4 times in one event should be 0.005% - one in 20,000.

If we have roughly 20,000 players per event, and 1% of the playerbase are active forum contributors, we'd expect to see a contributor point out this remarkable random occurrence on average once in every 100 events. Given the occasions where people get into the spirit of posting this stuff, where we see multiple people drawing 4 of the same opponent in the same event, and not a soul claiming to have faced 30 unique opponents (which should be happening to 86% of people, if random), I for one think we can safely eliminate the null hypothesis.

buscemi · February 2017

Volrak wrote:

buscemi wrote:

I think it's highly relevant when talking about shonky RNGs in a game to consider other shonky RNGs in the same game.

I'm not concerned about the possibility that matchmaking biases imply systemic biases in all random game events, because underlying RNG library calls are simple and well tested (by thousands of apps), while any event-specific code (e.g. matchmaking) by definition won't affect other types of RNG result (e.g. card drops).

The assumption that the dev team have used standard library functions to handle randomness without substantial modification is a very large one. The library functions do work well, but their results may have been fed through any number of other functions, and the random numbers generated may have become tainted on their way to influencing graphical elements on the screen. It's even possible the library functions themselves may not have been used. Don't laugh. I've worked for places that have done that.

Volrak wrote:

buscemi wrote:

We don't *know* that matchmaking has a strong non-random component

I for one think we can safely eliminate the null hypothesis.

However, if the *intention* of the dev team was to make matchmaking random, and whilst in attempting to do this they somehow ended up with a system that is clearly non-random in practice, then it's fair to assume that any of the systems in the game which were intended to be random can be skewed to any degree.

Hibernum's handling of player decks illustrates that they are a team that don't think that pure random is 'random enough' to eliminate perfectly normal attributes of randomness like clumpyness. Doesn't it seem at least plausible that a team which seeks to reduce clumpyness in card draw would also seek to reduce it in cascades, an area of the game which players frequently complain about? Doesn't Hibernum's past history also suggest the possibility that if they'd got the code wrong, and it did the opposite of what it was supposed to, they'd leave it in place because it wasn't a priority and get on with solving a different problem?

Volrak · February 2017

buscemi wrote:

The assumption that the dev team have used standard library functions to handle randomness without substantial modification is a very large one. The library functions do work well, but their results may have been fed through any number of other functions, and the random numbers generated may have become tainted on their way to influencing graphical elements on the screen. It's even possible the library functions themselves may not have been used. Don't laugh. I've worked for places that have done that.

Yes, it's a possibility. But if some intermediate function is causing the bias in matchmaking, and the same intermediate function is being used for other random events like gem drops, then I'd expect to see the same bias in gem drops. Applying the huge matchmaking biases to gem drops would mean for every 30 gems that drop, we'd expect to get on average something like 15 blue gems, 10 green gems, and one or two of some other colours, if you're lucky. This clearly isn't happening.

buscemi wrote:

However, if the *intention* of the dev team was to make matchmaking random, and whilst in attempting to do this they somehow ended up with a system that is clearly non-random in practice, then it's fair to assume that any of the systems in the game which were intended to be random can be skewed to any degree.

Indeed - this is similar to one of my comments in my previous post

buscemi wrote:

Hibernum's handling of player decks illustrates that they are a team that don't think that pure random is 'random enough' to eliminate perfectly normal attributes of randomness like clumpyness. Doesn't it seem at least plausible that a team which seeks to reduce clumpyness in card draw would also seek to reduce it in cascades, an area of the game which players frequently complain about? Doesn't Hibernum's past history also suggest the possibility that if they'd got the code wrong, and it did the opposite of what it was supposed to, they'd leave it in place because it wasn't a priority and get on with solving a different problem?

But on the other hand, pure random does appear to be 'random enough' for the devs when it comes to rarities of card drops, even though adding a pity timer on mythics would reward only the unluckiest players and remove a whole lot of feel-bad.

Like I said - I don't think there's a systemic code issue linking matchmaking to other random events, but I in no way rule out just plain bugginess (I suppose you could call this possibility a systemic quality issue). In general though, without evidence to the contrary, I'll tend to assume the simplest explanation - which for cascades, is that gems are produced with uniform randomness. If anyone wants to collect the data I'd be happy to take a closer look.

Emrakul's Corruption: AI Cascades

Comments

Categories