Implement event chance smoothing on EV and SH

[byrel]

Inspired by VeryAngryFelid's complaints about spikeyness in another thread and context.

EV and SH builds have a harder time with intuitive threat assessment than AC builds. You may be able to tab through 3 orc warriors while taking no damage, while the fourth high-rolls twice and nearly kills you. It's a natural feature of random events; you get runs of hits and misses far longer and more often than feels statistically likely. This leads to higher learning curve and burning consumables in not-very-threatening fights somewhat more often. Burning consumables is good, but it feels better when it's on opponents that are innately scary, and not due to just a sequence of bad rolls (and it's easier to balance the game when a large portion of combos don't have 10% fewer consumables due to more susceptibility to randomness).

The most straightforward solution to the remarkably long runs that random variables are prone to is an event-smoothing algorithm like MOBAs use for crits. Basically, instead of actually having a fixed chance to crit on every attack (which leads to multiple crits in a row and huge dry stretches more than feels right to people) they have a smaller-than-nominal chance to crit, which increases every time you don't crit (and resets when you do.) By selecting the initial event-chance and the non-event-increment appropriately, you can get the expected event rate to be the same as your uniform distribution. This sort of thing is also done for some loot crates (like Hearthstone card packs).

For the more visual taverners, here's the effect of the dota smoothing algorithm for a 25% event chance:


Obviously, this 'event state' would need to be reset for each monster that attacks you, to avoid stupid things like toting a pet kobold around with you to tank till you get a few consecutive misses before every fight.

I think this would probably improve the game feel if implemented for EV and SH, and I'd be willing to throw together a pull request if there was some interest.

[duvessa]

Oh great, more invisible state information to keep track of in fights.

[tealizard]

Spiky, swingy combat is good. Increasing the predictability of combat is bad.

[TheMeInTeam]

Spiky, swingy combat is good. Increasing the predictability of combat is bad.

On the contrary, most player actions by winning players are done to make the anticipated outcome and actual outcome match. This is attainable in most cases.

That said, I'm not convinced the game needs OP's suggestion. The risk factor is part of the drawback to going lighter and relying on spells. Though at least we're not being given % values that are literally inaccurate like some of the fire emblem games.

[tealizard]

Spiky, swingy combat is good. Increasing the predictability of combat is bad.

On the contrary, most player actions by winning players are done to make the anticipated outcome and actual outcome match. This is attainable in most cases.

lol, what the hell?

[TheMeInTeam]

Spiky, swingy combat is good. Increasing the predictability of combat is bad.

On the contrary, most player actions by winning players are done to make the anticipated outcome and actual outcome match. This is attainable in most cases.

lol, what the hell?

I'm refuting "lolrandom is good" without supportive basis.

If "lolrandom is good" held, we should not expect good players to perform much better than bad players at crawl. The logical conclusion of maxed unpredictability is that player decisions can't reasonably trace causally to future experiences.

If you're saying that's a good thing, it would be interesting to see a coherent basis for why that is.

[tealizard]

If "lolrandom is good" held, we should not expect good players to perform much better than bad players at crawl. The logical conclusion of maxed unpredictability is that player decisions can't reasonably trace causally to future experiences.

This does not follow at all. I did not say anything about maximizing unpredictability either.

[TheMeInTeam]

If "lolrandom is good" held, we should not expect good players to perform much better than bad players at crawl. The logical conclusion of maxed unpredictability is that player decisions can't reasonably trace causally to future experiences.

This does not follow at all. I did not say anything about maximizing unpredictability either.

You "Spiky, swingy combat is good" without qualifiers. If this isn't an statement in favor of unpredictability, what were you trying to establish?

[tealizard]

It is a statement about crawl, made in a comparative context presented by the OP's suggestion. The OP suggests that combat is too spiky, I say otherwise. I also say moving crawl in the direction of less spiky combat would make crawl worse.

[byrel]

My claim is that crawl is differentially spiky between EV/SH heavy characters and AC heavy. If you think it should be equalized the other way, remove GDR.

[tealizard]

Okay, well, as to that point: It seems to me that having different ways to build one's defenses that have various advantages and disadvantages is the whole point of having three different defense stats.

[svendre]

Okay, well, as to that point: It seems to me that having different ways to build one's defenses that have various advantages and disadvantages is the whole point of having three different defense stats.

Having different ways to build defenses is good. Having one of them be quite a bit superior to the other methods is questionable. I think that's what is being said here, and I agree. AC is superior to EV and SH overall. It allows you to predict incoming damage more accurately, which equates to survivability in the long term. I think that if the notion is that all three are somehow equal, then this thread makes no sense from the start.

If we can come to a consensus that AC is superior overall (and I'm sure some people haven't figured it out yet), uncommon exceptions excluded, then it's a discussion about how EV/SH can be enhanced or how AC can be nerfed. As was pointed out already, EV/SH have at least some advantage because they permit spell casting more easily, but I think they still fall short for a myriad of reasons. If anything, that wearing heavy armour worses spell casting has just served to make spell casting quite a bit worse than melee until armour is well trained and you find dragon scales.

I don't particularly like the solution of adding a layer of invisible state information as duvessa has stated quite well, and possibly one of the few times we agree about anything.

It seems a bit unfair for people still struggling to figure out why they die all the time, and how the heck are they supposed to progress in the game because they play too many dex based stealthy dodgy stabbers with powerful offense, glass cannons (and then they get paralyzed and learn the hard way) - that the skills presented for the same amount of XP should not be more balanced towards reliability. I've been taunted in the past for saying things like "just get 80 AC on a gargoyle, with negative EV and you're fine", but more and more people are coming to realize that you can win with only AC (and STR), and that winning without AC (and just tons of DEX) is much harder.

I don't see a lot of solutions, except to perhaps make each defensive strategy be more needed. The only thing I recall coming up with (half-baked) to address all this begins with the base stats. It would involve a lot of changes, removing elemental resists out of gear and putting them into dex, or a new stat of agility and reducing the amount of elemental damage that AC alone can block. So the heavy plate mail becomes great at stopping clubs, but it hurts even worse when blasted by lightning, with more than a resist making the difference. So a player could choose to be well protected by elemental/spell attacks by going dodgy, or well protected against brute force in heavy armour, but you would need some of both for the complete defense. I think it'd be great if there were more depth given about damage types as well. For example, in Unreal World (a survival game), you can wear wool stuff to dampen blunt weapons, but it won't help against arrows - and conversely, mail will stop blades but isn't as good at stopping blunt weapons, but I digress from the main point here.

This state was worsened also with the changes in the name of simplicity for STR to be the only factor for for damage and DEX, accuracy for all weapons, instead of differentiating a dagger from a great mace. The change didn't directly impact defenses obviously, but it did indirectly by affecting decisions about how to distribute stats. Maybe the old str weighting wasn't having a large impact, but it had some, and it should have a much greater impact, not less. You don't need much DEX to hit stuff when your skills are high, but you need more STR than ever to do the damage.

[tealizard]

I would not claim that crawl's AC/EV/SH system is anything amazing, but it is okay and none of the arguments in this thread or alternative proposals show any promise of improving on it.

[tasonir]

Half the people want smooth combat, the other half think that would be too easy. There's a solution for both camps:

Just put out a fork where you get the absolute worst result at all times. Your attacks hit for zero, they hit for their max damage all the time. It'll make everyone happy.

[svendre]

I would not claim that crawl's AC/EV/SH system is anything amazing, but it is okay and none of the arguments in this thread or alternative proposals show any promise of improving on it.

It is OK, I don't hate crawl. I thought this forum is for people who like crawl, to present ideas to potentially improve it. If you disagree with ideas presented, and you're going to be vocal about it, it would be nice if at least you state some reasons why rather than simply saying that ideas presented have no merit. I know you are very capable of writing more thoughtful responses.

[svendre]

Half the people want smooth combat, the other half think that would be too easy. There's a solution for both camps:

Just put out a fork where you get the absolute worst result at all times. Your attacks hit for zero, they hit for their max damage all the time. It'll make everyone happy.

I'm in a different camp. I'm less concerned with how spikey the game is overall (though I think it shouldn't be on the more spikey side) than I am with the skill and stat investments being more balanced among themselves.

[tealizard]

It is OK, I don't hate crawl. I thought this forum is for people who like crawl, to present ideas to potentially improve it. If you disagree with ideas presented, and you're going to be vocal about it, it would be nice if at least you state some reasons why rather than simply saying that ideas presented have no merit. I know you are very capable of writing more thoughtful responses.

I'd be happy to have a more substantive discussion of these issues, but I don't see the basis for that in the ideas I'm seeing. You have no business reproaching me for not engaging sufficiently with your ideas when there are plenty of other people in the position to comment who don't even bother.

[petercordia]

I like swingy, spiky combat, because it makes you think.

I agree with Svendre that having different kinds of defence which work in different ways is fun, and that AC might be too powerful compared with EV and SH. Since AC is more reliable, and is unaffected by paralyses and confusion, it might be fair for AC to be - on average - less effective than EV or SH.
A relatively simple way to balance this out (if that is desirable) would be to increase the players EV and SH with some factor and decrease the AC with a similar factor. eg:

EV_new = EV_old * 1.2
SH_new = SH_old * 1.2
AC_new = AC_old / 1.2

Since EV, SH and AC are already calculated with formula which return non-integer numbers, I don't expect rounding would be a problem.

[byrel]

Duvessa, I'm not sure I see your objection. While it's true that this adds state, in what way would tracking that state provide an advantage? If it doesn't, then you'd have no incentive to track it and it shouldn't reduce optimal player burden.

[pedritolo]

I suppose it lets you game the system by leaving dangerous fights aside, then search for an easy monster (like a goblin) that you parked away somewhere, taking off your armour and shield, and wait until the goblin hits you through your ev, which might indicate that your next ev hit will on average take a while to show up. Personally I'd never go through all this trouble, but someone else might.

[sanka]

I suppose it lets you game the system by leaving dangerous fights aside, then search for an easy monster (like a goblin) that you parked away somewhere, taking off your armour and shield, and wait until the goblin hits you through your ev, which might indicate that your next ev hit will on average take a while to show up. Personally I'd never go through all this trouble, but someone else might.

Obviously, this 'event state' would need to be reset for each monster that attacks you

[pedritolo]

ah gotcha

[Siegurt]

I suppose it lets you game the system by leaving dangerous fights aside, then search for an easy monster (like a goblin) that you parked away somewhere, taking off your armour and shield, and wait until the goblin hits you through your ev, which might indicate that your next ev hit will on average take a while to show up. Personally I'd never go through all this trouble, but someone else might.

Obviously, this 'event state' would need to be reset for each monster that attacks you

It would still be optimal to track your state vs. each monster. If you're lower on HP it'd be more optimal to run away when you have higher odds of being hit, it'd be more optimal to take your lumps when your HP is higher.

[byrel]

It would still be optimal to track your state vs. each monster. If you're lower on HP it'd be more optimal to run away when you have higher odds of being hit, it'd be more optimal to take your lumps when your HP is higher.

I don't think you've thought through this scenario fully. Suppose you do track this state perfectly, and, for an initial condition, you're a 50HP melee dude fighting an ogre.
In his first swings, you deal half his HP.
In Scenario A, you get missed twice, then hit for 20 damage.
In Scenario B, you get missed three times.

Would your knowledge of the state in scenario B make you more likely to run (because you know a hit is more likely than when you engaged) than in scenario A where you're now in one-shot range of the ogre?

Counting misses, and knowing your odds to get hit are going up, doesn't actually impact decision making in this case. If you had enough HP to safely engage the ogre, the fact that he's more likely to successfully hit you soon isn't going to change your decision; you already planned to tank a hit, and you've dealt damage to him. Him missing you is still strictly better for you than him hitting you, so if he's more likely to hit you it's because you've been luckier than your expected value which made you engage in the first place.

Can you think of a concrete scenario like this where perfect knowledge of the state would change the optimal decision?

[duvessa]

Can you think of a concrete scenario like this where perfect knowledge of the state would change the optimal decision?

There are two ogres. You can attack one of the ogres. It is better to attack the ogre that is more likely to hit you. With state tracking, you know which ogre this is; without state tracking you do not.

[tealizard]

Ah, but you see... it may be impossible to determine which of the two ogres has hit or missed on recent turns!

It is not necessary to get into specific examples to address the issue here. The distribution of probabilities versus trial without "event smoothing" shown is actually the most unpredictable possible way to do it subject to the constraint that the average number of trials needed to see the event is 4. Therefore you will get more predictable combat using this "event smoothing" business. Now how do you actually take advantage of this greater predictability? The answer is by tracking the state information used produce the difference between smoothed and unsmoothed. If the claim is that actually the additional predictability makes no difference in any particular case, then I conclude the effect is too small to be worth considering in the first place. To the extent it does make a difference though, it does so by moving away from the least predictable way to do it, which is bad in my view.

[Hellmonk]

and here I was going to give an example about teleporting away from a dangerous monster that either missed or hit me, saving my game, and then running into it again hundreds of turns and several days later

[VeryAngryFelid]

Is it correct that the change would buff high AC characrers and it would be better to be hit for 0-1 damage than missed?

[ion_frigate]

Invisible states like this seem pretty much unworkable. Crawl is open-source. Therefore it's possible to build a tool that will track these states, given that they depend on information that is at some point available to the player. I can easily imagine doing this programmatically by directly reading terminal output, building a tool that will keep track of monsters in sight and how many times they've hit/missed, so you can know which ones are likely to get lucky soon - it might not be perfect, but it'll be a hell of a lot better than not knowing the monsters' luck state at all.

If such a tool is built, it will be used in tournaments. Then to avoid unbalancing the tournament, you have to make the tool available in-game to everyone, i.e. adding the monster's 'crit state' to its xv description. Great, another stupid piece of information to keep track of, one that leads to decisions that are more or less no-brainers (attack the ogre that's most likely to get 35 damage with his club next turn).

The OP is coming from experience playing MOBAs. The thing about those games is that they're closed-source, they're real-time, and their competitive scene takes place in highly controlled circumstances. That means that skilled DoTA or LoL players *can't* use any tools to track luck states: they don't have time to read such a tool, it would be banned in tournaments, and even building it would require hundreds of hours of experimentation rather than a few hours of coding. None of those things are true for an open-source, turn-based game whose competitive scene consists of semi-formal tournaments where there is absolutely no client-side control.

Also, I would like to point out there are plenty of other ways of addressing the OP's original issue, i.e. spikiness in EV/SH builds. The most basic way is to just roll more dice in general. Rework damage formulas so you don't have the primary component of damage being a big 1d30 dice roll.

[tealizard]

So suppose it's not invisible. You could certainly imagine some visual indicator of the sort of combat state under discussion. At some point you have to make an argument that crawl combat is too unpredictable as it stands. That's going to be a hard in a game with an ideal winrate nearly 100%.

As far as the provenance of the idea, as I see it the OP is more or less: "In DoTA/LoL/whatever, they have 'random crits,' which is to say a random element in a game based on timing, real time execution and decision making. As you might guess, the hot pocket-deprived 13 year olds who play this shit get kind of pissed if they get hit three times in a row with a 25% crit attack or hit another 13 year old 10 times in a row without a 'random crit,' so they came up with a mechanic that keeps that from happening. Now imagine there was a crawl player who thought like the LoL kid..." I just have to say, maybe, just maybe, this idea of a "random crit" isn't a good fit for that kind of game and that's why they have to throw in this additional, de-randomizing element.

As for crawl, it's fundamentally a game of chance. There's no real time element. The randomization is the only thing that generates action. De-randomizing combat would detract from what tension is there.

[VeryAngryFelid]

Inspired by VeryAngryFelid's complaints about spikeyness in another thread and context.

Note that I complained about quite the opposite thing.
Imagine you have damage rolls like 2, 1, 2, 4, 1, 2, 2, 100, 0, 1, 2, 1, 1, 0, 0, 79, 1, 4, 5, 1, 99. I suggested to somehow reduce number of low damage attacks between high damage attacks, not increase.
I don't have any problems with 2, 100, 80, 3, 60, 75, 99, 150, actually this is what happens when we reduce number of low damage attacks.

[tealizard]

If you remove low/no damage attacks and keep high damage ones, you will increase the average damage per round, which will disrupt the balance between combat movement and other combat actions. That will lead to more static combat. The current balance is pretty good, so I would not want to mess with it too much.

The most important variables to control are average damage and maximum damage. With those parameters in mind, the distributions used in crawl combat are not optimal, so there is room for improvement. The optimal distributions are heavy on low damage outcomes, but a bit lighter on the very high and very low ends.

[VeryAngryFelid]

If you remove low/no damage attacks and keep high damage ones, you will increase the average damage per round, which will disrupt the balance between combat movement and other combat actions. That will lead to more static combat. The current balance is pretty good, so I would not want to mess with it too much.

It looks like we have different understanding of "static". For me static is when you attack a monster 5 times, the monster attacks you 5 times and both are still at full HP. Actually if you want to have situation when every action counts, you should be proponent of reducing number of low damage attacks and increasing both average and max damage. It is both weird and bad when max damage is dozen of times higher than average damage and median damage is 20-30% of average damage.

[byrel]

Can you think of a concrete scenario like this where perfect knowledge of the state would change the optimal decision?

There are two ogres. You can attack one of the ogres. It is better to attack the ogre that is more likely to hit you. With state tracking, you know which ogre this is; without state tracking you do not.

You're right, and I can't think of a way of eliminating this.

I withdraw the suggestion; it's terrible unless the state literally doesn't matter to gameplay.

[svendre]

If you remove low/no damage attacks and keep high damage ones, you will increase the average damage per round, which will disrupt the balance between combat movement and other combat actions. That will lead to more static combat. The current balance is pretty good, so I would not want to mess with it too much.

The most important variables to control are average damage and maximum damage. With those parameters in mind, the distributions used in crawl combat are not optimal, so there is room for improvement. The optimal distributions are heavy on low damage outcomes, but a bit lighter on the very high and very low ends.

This made me think of Battle for Wesnoth. I love the game dearly, but it is brutally annoying when the results just barely diverge from probability, which is even displayed beforehand. In just a mere few turns, because of a sour outcome of rolls you can go from doing wonderfully to completely screwed. It doesn't have the layer of "a bunch of low/no damage attacks" in the mix. I know it has been grating on a number of people besides myself, and I think there might even have been some option added to the game to soften it although I don't recall for sure or the details offhand.

I think these low numbers and zeros combined with the possibility of large hits is a good balance between the annoyance of spikes and annoyance from lack of suspense, while keeping it random.

[svendre]

It is OK, I don't hate crawl. I thought this forum is for people who like crawl, to present ideas to potentially improve it. If you disagree with ideas presented, and you're going to be vocal about it, it would be nice if at least you state some reasons why rather than simply saying that ideas presented have no merit. I know you are very capable of writing more thoughtful responses.

I'd be happy to have a more substantive discussion of these issues, but I don't see the basis for that in the ideas I'm seeing. You have no business reproaching me for not engaging sufficiently with your ideas when there are plenty of other people in the position to comment who don't even bother.

I wasn't implying that I think you should have to care about any ideas presented. What I meant was that if, and *only* if you are going to make a statement that no ideas presented have merit, that it would be polite to also explain why, otherwise just not say anything. It's no big deal to me if we simply disagree on the protocol. I didn't see a huge mess of thoughtless ideas, which is the only case where I may feel I want to blurt out something like this all sucks. If there was some gigantic pile of tedious ideas too ridiculous to respond to, I probably wouldn't be all that concerned about informing everyone else that they were all bad ideas, because wouldn't they be able to see that for themselves in this case?

Finally, why would I "reproach" people in a position to comment, who simply did not comment? If a person says nothing positive or negative at all, it's a non-factor and does not compare with a person who unilaterally dismisses all ideas.