Solutions to luck and their problems
Moderator: Forum Moderators
Solutions to luck and their problems
Hello.
There are a lot of topics ranting about luck. Instead of complaining about I want to discuss the different ways you could hypothetically modify wesnoth to reduce the likelihood of luck ruining somebody's game while keeping the core game intact.Note:
- The problem I'm discussing is that many people appear to dislike how luck can heavily influence games. Luck itself is not a problem that needs to be solved, how it apparantly negatively impacts many people is.
- I do not think the core game itself should be changed, just that perhaps alternatives can be researched. Optimally some could be implemented in the form of optional add-ons.
- I don't want to discuss how necessary/good/bad the luck as it is now is., there are plenty of topics about that.
- It is very well possible that the conclusion of this thread will be that there is no good solution.
Solution 1: Pseudo Random Number Generator
Technically all RNG's implemented in computers are Pseudo RNG. What I mean with this solution are RNG's designed to prevent long streaks of very bad/good luck.
There are several ways to implement this:
- whenever an attack hits, reduce the chance of the next attack hitting, whenever an attack misses, reduce the chance of the next attack missing, etc. Compound chances for consuctive misses/hits.
- for each player, generate the next n random numbers they will use when attacking, if any subsequence of k numbers contains more than l numbers higher than X or lower than (1-X), throw away the series and generate a new one (don't test the new one). Example values: generate 20 numbers, if in any subsequence of 9 numbers 6 or more are all higher than 80% or all lower than 20%, regenerate everything.
- for each player use separate rng's for their attack rolls. Every time you make a random number, modify it by the radio of sqrt((average expected)/(average past n numbers)). Example, if a player's past 4 numbers average 0.4 and the next number generated is a 0.6, multiply 0.6 with (0.5/0.4).
- etc
Sadly, there are many problems with this:
- it is inevitable that the player would be able to game and/or predict the system. Good players would keep track of their rng luck and attack in an order that makes sure their most important attacks have an extra high chance of hitting.
- how better the pseudo rng is at preventing bad luck streaks, the worse the above problem becomes.
- the implementation of this is not that straightforward.
- when bad luck streaks do occur despite the pseudo rng (and they must), the frustration of the player might very well be even moreso than normal considering he'd know it was even less likely.
- reduces how important it is to plan for all possible outcomes
As such, I personally don't think this is a good idea.
Solution 2: no luck
This "solution" is basically to make a different game, that just happens to have exactly the same units/campaigns/etc as wesnoth. Obviously you wouldn't change wesnoth itself to work like this, but rather use these ideas as optional add-ons.
There are again several ways to implement this:
- All attacks hit, but their damage is reduced by the chance they would have missed. So a 10-2 attack against an enemy on a 60% defense tile would always deal 8 damage. I believe this is implemented already.
- When attacking, compute the expected damage done/taken (i.e. consider all possible outcomes and their likelihood, take a weighted average) and deal/take that amount of damage.
Problems:
- well, it's not the same game anymore.
- completely destroys how important it is to consider all possible scenario's.
- it results in very boring gameplay.
This is just not viable in my opinion.
Solution 3: hybrid systems
Examples:
- as the above no luck methods, except when an attack could result in a kill. Ie if you attack 40 hp unit with a 10-2 attack at 60% hit chance you'll always deal 12 damage. If you attack the same unit at 20 hp, combat goes as normal and the unit takes 0, 10 or 20 damage.
- as the above no luck methods, but with some randomness in the damage done/taken. I.e. when attacking with a 10-2 attack with a 60% hit chance you'll do between 8 and 16 damage randomly (instead of 12 always).
Problems:
- the first is just extremely confusing to the user because of how inconsistent it is.
- the second is a bit unelegant and has both the problems of the current rng and the no luck possibility.
Personally the second solution using the "deal the expected damage" would be something I'd like to try, but it does seem a bit silly to me to change which kind of luck is used.
Solution 4: no exceptional luck
Consider all possible damage done outcomes of an attack, remove any outcomes that are less likely than 5%. So it's possible to miss all 3 attacks at 60% but not to miss all 4 and it is not possible to miss all 3 attacks at 70% hit chance. Similarly, you can no longer hit all 4 attacks against an opponent with 60% defense or all 3 attacks against an opponent with 70% defense. Can either be implemented by just taking a new random sequence when a sequence would result in this, or by computing the likelood of the leftover attacks considering the attacks with <5% cannot be picked. Obviously an exception would be made for ulfzerkers.
This is a combination of solution 1 and 2. While solution 1 reduces the likelihood of exceptional sequences and hence allows players to cheat the system, this solution simply disallows exceptional sequences. Obviously players can also cheat this system (gonna attack that 1hp rogue with an attack that has 4 hits so I'm 100% sure one hits), but since it is much more transparent this is less of a problem. In addition, because unlike solution 1 this does not propagate between units or turns the ways you can cheat with this are slightly more limited.
Problems:
- You no longer have to plan for all possible outcomes.
- It takes quite a bit of calculation to know which patterns are impossible, so it violates KISS.
- Not that easy to implement.
I'd like to try this out, especially in multiplayer, but in the end I fear this complicates matters so much that there's little benefit to be had.
Solution 5: extra abilities/traits that influence luck
This solution also somewhat solves something I also personally consider to be a problem: namely that units are often too similar. The idea is to add abilities and/or weapon abilities that interact with luck.
Examples:
- Ability "nimble", can never be hit 3 times in a row while defending. Balance with -1/2 hp.
- Weapon ability "feint", can never miss 3 attacks in a row while attacking or hit 3 attacks in a row while attacking.
- Ability "disengage", if you miss all your attacks while attacking, regain 2 movement points. Balance with -1 damage. Alternatively, does not consume leftover movement points when all your attacks miss.
- Weapon ability "reach", deal 1 damage each time you miss. Balance with -1 damage.
- Ability "frustration", each time you miss, your next attack is 10% more likely to hit, each time you hit, your next attack is 10% more likely to miss. Does not compound and is reset after every fight.
- ABility "guardian", while defending, the first time you are attacked by an attack that could kill you, you are certain to evade that attack (it could have missed without this ability, but now you are sure). Can activate once a day, i.e. every 6 turns. Balance with -5 hp.
- Weapon ability "finisher", while attacking, if your last attack could finish the enemy, it is twice as likely to hit. Only given to weapons with 4+ attacks (preferably 5), balanced with -1 damage.
- etc, etc...
Obviously not all units would have traits like this, but rather you'd give it to units that are too similar to other units to differentiate and/or promotions that are underused. Like poacher, javelineer, elven fighter, lancer, elvish sorceress, etc. Optimally each race should have some abilities like this. Note that statistically these abilites are all pretty weak, especially the first 2.
Problems:
- hard to implement.
- changes balancing, not always straightforward how to balance it.
- complicates the game (this can be seen as a good thing as well).
- doesn't really solve the problem.
I'd like this very much and feel this would be an ideal compromise, but fear it would be difficult to implement.
Conclusion
None of the solutions seem ideal to me, but some would be interesting to try and many are certainly worth a discussion. I feel implementing this is more the responsibility of the modding community than the devs themselves. Although I would like to try implementing these myself, I'm not sure if I have the time and/or know-how to do so.
This topic could have been posted in the ideas subforum as well, I'm not sure where it fits better.
There are a lot of topics ranting about luck. Instead of complaining about I want to discuss the different ways you could hypothetically modify wesnoth to reduce the likelihood of luck ruining somebody's game while keeping the core game intact.Note:
- The problem I'm discussing is that many people appear to dislike how luck can heavily influence games. Luck itself is not a problem that needs to be solved, how it apparantly negatively impacts many people is.
- I do not think the core game itself should be changed, just that perhaps alternatives can be researched. Optimally some could be implemented in the form of optional add-ons.
- I don't want to discuss how necessary/good/bad the luck as it is now is., there are plenty of topics about that.
- It is very well possible that the conclusion of this thread will be that there is no good solution.
Solution 1: Pseudo Random Number Generator
Technically all RNG's implemented in computers are Pseudo RNG. What I mean with this solution are RNG's designed to prevent long streaks of very bad/good luck.
There are several ways to implement this:
- whenever an attack hits, reduce the chance of the next attack hitting, whenever an attack misses, reduce the chance of the next attack missing, etc. Compound chances for consuctive misses/hits.
- for each player, generate the next n random numbers they will use when attacking, if any subsequence of k numbers contains more than l numbers higher than X or lower than (1-X), throw away the series and generate a new one (don't test the new one). Example values: generate 20 numbers, if in any subsequence of 9 numbers 6 or more are all higher than 80% or all lower than 20%, regenerate everything.
- for each player use separate rng's for their attack rolls. Every time you make a random number, modify it by the radio of sqrt((average expected)/(average past n numbers)). Example, if a player's past 4 numbers average 0.4 and the next number generated is a 0.6, multiply 0.6 with (0.5/0.4).
- etc
Sadly, there are many problems with this:
- it is inevitable that the player would be able to game and/or predict the system. Good players would keep track of their rng luck and attack in an order that makes sure their most important attacks have an extra high chance of hitting.
- how better the pseudo rng is at preventing bad luck streaks, the worse the above problem becomes.
- the implementation of this is not that straightforward.
- when bad luck streaks do occur despite the pseudo rng (and they must), the frustration of the player might very well be even moreso than normal considering he'd know it was even less likely.
- reduces how important it is to plan for all possible outcomes
As such, I personally don't think this is a good idea.
Solution 2: no luck
This "solution" is basically to make a different game, that just happens to have exactly the same units/campaigns/etc as wesnoth. Obviously you wouldn't change wesnoth itself to work like this, but rather use these ideas as optional add-ons.
There are again several ways to implement this:
- All attacks hit, but their damage is reduced by the chance they would have missed. So a 10-2 attack against an enemy on a 60% defense tile would always deal 8 damage. I believe this is implemented already.
- When attacking, compute the expected damage done/taken (i.e. consider all possible outcomes and their likelihood, take a weighted average) and deal/take that amount of damage.
Problems:
- well, it's not the same game anymore.
- completely destroys how important it is to consider all possible scenario's.
- it results in very boring gameplay.
This is just not viable in my opinion.
Solution 3: hybrid systems
Examples:
- as the above no luck methods, except when an attack could result in a kill. Ie if you attack 40 hp unit with a 10-2 attack at 60% hit chance you'll always deal 12 damage. If you attack the same unit at 20 hp, combat goes as normal and the unit takes 0, 10 or 20 damage.
- as the above no luck methods, but with some randomness in the damage done/taken. I.e. when attacking with a 10-2 attack with a 60% hit chance you'll do between 8 and 16 damage randomly (instead of 12 always).
Problems:
- the first is just extremely confusing to the user because of how inconsistent it is.
- the second is a bit unelegant and has both the problems of the current rng and the no luck possibility.
Personally the second solution using the "deal the expected damage" would be something I'd like to try, but it does seem a bit silly to me to change which kind of luck is used.
Solution 4: no exceptional luck
Consider all possible damage done outcomes of an attack, remove any outcomes that are less likely than 5%. So it's possible to miss all 3 attacks at 60% but not to miss all 4 and it is not possible to miss all 3 attacks at 70% hit chance. Similarly, you can no longer hit all 4 attacks against an opponent with 60% defense or all 3 attacks against an opponent with 70% defense. Can either be implemented by just taking a new random sequence when a sequence would result in this, or by computing the likelood of the leftover attacks considering the attacks with <5% cannot be picked. Obviously an exception would be made for ulfzerkers.
This is a combination of solution 1 and 2. While solution 1 reduces the likelihood of exceptional sequences and hence allows players to cheat the system, this solution simply disallows exceptional sequences. Obviously players can also cheat this system (gonna attack that 1hp rogue with an attack that has 4 hits so I'm 100% sure one hits), but since it is much more transparent this is less of a problem. In addition, because unlike solution 1 this does not propagate between units or turns the ways you can cheat with this are slightly more limited.
Problems:
- You no longer have to plan for all possible outcomes.
- It takes quite a bit of calculation to know which patterns are impossible, so it violates KISS.
- Not that easy to implement.
I'd like to try this out, especially in multiplayer, but in the end I fear this complicates matters so much that there's little benefit to be had.
Solution 5: extra abilities/traits that influence luck
This solution also somewhat solves something I also personally consider to be a problem: namely that units are often too similar. The idea is to add abilities and/or weapon abilities that interact with luck.
Examples:
- Ability "nimble", can never be hit 3 times in a row while defending. Balance with -1/2 hp.
- Weapon ability "feint", can never miss 3 attacks in a row while attacking or hit 3 attacks in a row while attacking.
- Ability "disengage", if you miss all your attacks while attacking, regain 2 movement points. Balance with -1 damage. Alternatively, does not consume leftover movement points when all your attacks miss.
- Weapon ability "reach", deal 1 damage each time you miss. Balance with -1 damage.
- Ability "frustration", each time you miss, your next attack is 10% more likely to hit, each time you hit, your next attack is 10% more likely to miss. Does not compound and is reset after every fight.
- ABility "guardian", while defending, the first time you are attacked by an attack that could kill you, you are certain to evade that attack (it could have missed without this ability, but now you are sure). Can activate once a day, i.e. every 6 turns. Balance with -5 hp.
- Weapon ability "finisher", while attacking, if your last attack could finish the enemy, it is twice as likely to hit. Only given to weapons with 4+ attacks (preferably 5), balanced with -1 damage.
- etc, etc...
Obviously not all units would have traits like this, but rather you'd give it to units that are too similar to other units to differentiate and/or promotions that are underused. Like poacher, javelineer, elven fighter, lancer, elvish sorceress, etc. Optimally each race should have some abilities like this. Note that statistically these abilites are all pretty weak, especially the first 2.
Problems:
- hard to implement.
- changes balancing, not always straightforward how to balance it.
- complicates the game (this can be seen as a good thing as well).
- doesn't really solve the problem.
I'd like this very much and feel this would be an ideal compromise, but fear it would be difficult to implement.
Conclusion
None of the solutions seem ideal to me, but some would be interesting to try and many are certainly worth a discussion. I feel implementing this is more the responsibility of the modding community than the devs themselves. Although I would like to try implementing these myself, I'm not sure if I have the time and/or know-how to do so.
This topic could have been posted in the ideas subforum as well, I'm not sure where it fits better.
Re: Solutions to luck and their problems
Solution 4 seems to have an additional problem: some units and factions would basically be exempt from the rule. If you play northerners, all but one of your units have at most 3 strikes. You can have all those grunts and trolls and archers and wolf riders attack that enemy unit with 60% defense and miss all their strikes, because all attacks are considered individually and missing 3/3 against 60% defense is as likely as 6.4%. Meanwhile the enemy might have plenty of units with 4 strikes and is guaranteed to hit at least once with each one. Wherever you put the threshold, it seems that it'd sharply divide attack opportunities into "not guaranteed to hit" and "guaranteed to hit" and then the game would start to revolve around that, especially in situations where getting that one killing blow in is critical.
EDIT: Solution 1 is similar to the simple version that was the LessLuck Era. However, as discussed in the thread, the -10/+10% luck modifier I used was too simple and didn't work as intended. I never updated it to use a non-broken formula because IIRC it would have been a bit too tricky to pull off (at least at the time), but doing that would probably be my suggestion for a modern lessluck mod.
EDIT: Solution 1 is similar to the simple version that was the LessLuck Era. However, as discussed in the thread, the -10/+10% luck modifier I used was too simple and didn't work as intended. I never updated it to use a non-broken formula because IIRC it would have been a bit too tricky to pull off (at least at the time), but doing that would probably be my suggestion for a modern lessluck mod.
Re: Solutions to luck and their problems
3 strikes would already matter against opponents with either 70% defense or 30% defense (or less) though. Either way I'd consider it intentional of that system that grunts/thunderers/other units with few attacks keep their "all or nothing" approach (as they are not impacted by the change), yet that units with lots of attacks get emphasis on their relieability. If this is fair among all races is a different question entirely, but I don't think it has to be.zookeeper wrote:Solution 4 seems to have an additional problem: some units and factions would basically be exempt from the rule. If you play northerners, all but one of your units have at most 3 strikes. You can have all those grunts and trolls and archers and wolf riders attack that enemy unit with 60% defense and miss all their strikes, because all attacks are considered individually and missing 3/3 against 60% defense is as likely as 6.4%. Meanwhile the enemy might have plenty of units with 4 strikes and is guaranteed to hit at least once with each one. Wherever you put the threshold, it seems that it'd sharply divide attack opportunities into "not guaranteed to hit" and "guaranteed to hit" and then the game would start to revolve around that, especially in situations where getting that one killing blow in is critical.
EDIT: Solution 1 is similar to the simple version that was the LessLuck Era. However, as discussed in the thread, the -10/+10% luck modifier I used was too simple and didn't work as intended. I never updated it to use a non-broken formula because IIRC it would have been a bit too tricky to pull off (at least at the time), but doing that would probably be my suggestion for a modern lessluck mod.
I understand that there's a huge distinction between "very likely to kill" and "guaranteed to kill", but if I see I have 95%+ chance to kill (when this system comes in effect) I'd kinda assume I'd get the kill and it's exactly the scenario's in which that doesn't happen that cause the frustration and the problem with luck.
Re: Solutions to luck and their problems
Maybe. I just think that there's an inherent asymmetry regarding good and bad luck: if you know you won't be able to hit 4/4 anymore then that's not a big deal since getting that lucky doesn't really factor into your plans in the first place, whereas knowing that you're guaranteed to hit at least once is quite a game-changer. Even if most of the time you'll happily gamble on hitting at least 1/4 against 60%, I'd expect that knowing it for sure would give you a considerable advantage in many situations.Suira wrote:3 strikes would already matter against opponents with either 70% defense or 30% defense (or less) though. Either way I'd consider it intentional of that system that grunts/thunderers/other units with few attacks keep their "all or nothing" approach (as they are not impacted by the change), yet that units with lots of attacks get emphasis on their relieability. If this is fair among all races is a different question entirely, but I don't think it has to be.
Re: Solutions to luck and their problems
Another thought on the solution #1:
In my experience, when people talk about these auto-luck-balancing mechanisms, they usually have in mind something like you explained, which is pretty close to the Warcraft 3 "pseudorandom distribution". http://dota2.gamepedia.com/Pseudo-random_distribution
IIUC it's been used in an experimental version of wesnoth before, but I agree it has drawbacks like you also said.
However I think another approach would be to use a "low discrepancy" sequence, with a random shift, like explained here:
http://en.wikipedia.org/wiki/Quasi-Monte_Carlo_method
I actually first heard about this on these forums, it was in an AI thread though. The idea there was that an AI based on monte carlo simulations would use this instead of the RNG to have it converge faster.
But I guess you could also just make a mod where this is used entirely instead of the RNG in the actual game.
The point of it is that it's like a PRNG, but the numbers are ensured to be spaced evenly, in order to reduce variance without creating overall bias.
At some level it must be implicitly like the "pseudo" distribution you were talking about, where some/all attacks will have slightly different probabilities, biased based on earlier luck. However if it is too subtle / difficult to take advantage of in the game then it might be appropriate.
It is admittedly un-KISS :/ but perhaps worth considering for a mod / experimental BfW version.
In my experience, when people talk about these auto-luck-balancing mechanisms, they usually have in mind something like you explained, which is pretty close to the Warcraft 3 "pseudorandom distribution". http://dota2.gamepedia.com/Pseudo-random_distribution
IIUC it's been used in an experimental version of wesnoth before, but I agree it has drawbacks like you also said.
However I think another approach would be to use a "low discrepancy" sequence, with a random shift, like explained here:
http://en.wikipedia.org/wiki/Quasi-Monte_Carlo_method
I actually first heard about this on these forums, it was in an AI thread though. The idea there was that an AI based on monte carlo simulations would use this instead of the RNG to have it converge faster.
But I guess you could also just make a mod where this is used entirely instead of the RNG in the actual game.
The point of it is that it's like a PRNG, but the numbers are ensured to be spaced evenly, in order to reduce variance without creating overall bias.
At some level it must be implicitly like the "pseudo" distribution you were talking about, where some/all attacks will have slightly different probabilities, biased based on earlier luck. However if it is too subtle / difficult to take advantage of in the game then it might be appropriate.
It is admittedly un-KISS :/ but perhaps worth considering for a mod / experimental BfW version.
Re: Solutions to luck and their problems
Yes, you have a point. Do you think it would be possible/necessary to rebalance the units that benefit from this change? A little bit less hp, a little bit more experience needed to level up, etc.zookeeper wrote:Maybe. I just think that there's an inherent asymmetry regarding good and bad luck: if you know you won't be able to hit 4/4 anymore then that's not a big deal since getting that lucky doesn't really factor into your plans in the first place, whereas knowing that you're guaranteed to hit at least once is quite a game-changer. Even if most of the time you'll happily gamble on hitting at least 1/4 against 60%, I'd expect that knowing it for sure would give you a considerable advantage in many situations.Suira wrote:3 strikes would already matter against opponents with either 70% defense or 30% defense (or less) though. Either way I'd consider it intentional of that system that grunts/thunderers/other units with few attacks keep their "all or nothing" approach (as they are not impacted by the change), yet that units with lots of attacks get emphasis on their relieability. If this is fair among all races is a different question entirely, but I don't think it has to be.
Thinking about this, maybe it'd be better to implement solution 3 as solution 4 to make rebalancing easier. In other words, make a (weapon) ability that disallows possible outcomes with less than 5% chance (reword it into something understandable for the players) and rebalance the units you give it to as well as making sure every race has one. This way has the added benefit that you don't have to give it to mages, who might otherwise benefit too much (as the chance of 3 hits missing at 70% hit chance is <5% you would be guaranteed to hit at least one of them) and that it makes units feel more unique/different.
Yes, that's what I meant with pseudorandom number generator. However, the way it is implemented in those games is not possible here. In those games it uses a different pseudo RNG for each individual item/ability that requires it and the chance that the ability/item activates is always the same. In wesnoth you need the same rng for all your units and those all attack with different chances to hit, so it's just not viable to port it over 1-to-1.iceiceice wrote:Another thought on the solution #1:
In my experience, when people talk about these auto-luck-balancing mechanisms, they usually have in mind something like you explained, which is pretty close to the Warcraft 3 "pseudorandom distribution". http://dota2.gamepedia.com/Pseudo-random_distribution
IIUC it's been used in an experimental version of wesnoth before, but I agree it has drawbacks like you also said.
However I think another approach would be to use a "low discrepancy" sequence, with a random shift, like explained here:
http://en.wikipedia.org/wiki/Quasi-Monte_Carlo_method
I actually first heard about this on these forums, it was in an AI thread though. The idea there was that an AI based on monte carlo simulations would use this instead of the RNG to have it converge faster.
But I guess you could also just make a mod where this is used entirely instead of the RNG in the actual game.
The point of it is that it's like a PRNG, but the numbers are ensured to be spaced evenly, in order to reduce variance without creating overall bias.
At some level it must be implicitly like the "pseudo" distribution you were talking about, where some/all attacks will have slightly different probabilities, biased based on earlier luck. However if it is too subtle / difficult to take advantage of in the game then it might be appropriate.
It is admittedly un-KISS :/ but perhaps worth considering for a mod / experimental BfW version.
With regards to use a low discrepancy sequence with random shift: am I understanding it right that it works as follows:
- generate a very long sequence which has a low discrepancy. I.e. given any subsequence and any interval, the amount of numbers of the subsequence in that interval is around what you'd expect probabilistically.
- pick a random starting point in the sequence so users can't predict the next numbers
This sounds like a much better but more complex version of the 2nd way I described pseudo random number generators in solution 1. It'd be cool, but it still has the same problems as other pseudo random number generators. Simply because it is sufficient for the player to know that the pseudo random number generator is designed to reduce the chance of very luck/unlucky streaks to abuse the system. If the player does not understand the way the numbers are generated or not isn't that relevant, either way he'll adopt the Gambler's fallacy. http://en.wikipedia.org/wiki/Gambler%27s_fallacy
Wether the gambler's fallacy becoming true is a big issue or not is another question.
edit: I only just now realized that every and all balancing problems with solution 4 would also be present in any good implementation of solution 1. Zookeeper is right that the difference between 95% chance to hit at least once and 100% chance is quite significant, but if a pseudo random number generator were to be implemented that 95% chance to hit at least once would probably become something like 99.8% due to how pseudo rng's work, since that's essentially what they're designed to do. Hence, if a pseudo random number generator is ever implemented, we should be aware of a possible (small) impact on balance in favor of units with more strikes.
Considering how any change to luck is almost certain to influence strategy, I'm personally in favor of making it as obvious and clear to the players as possible. As such, I think that solution 4 is better than solution 1 and that solution 5 that is used to implement solution 4 would be optimal. In a certain way you could say that the "cheating" we talk about with regards to solution 1 will still exist (of course) in solution 4/5, but the difference is that because it is well documented as transparant that cheating becomes part of strategy. Manipulating a pseudo rng into making sure you kill an unit is (in my opinion) bad because it distracts from the normal gameplay, whereas if you correctly use an unit with the ability to always hit at least once in four strikes to make sure you kill an unit you are correctly using the resources the game gives you without you being distracted from the core gameplay. By making it part of gameplay in an interesting and simple way, you remove the problem.
Re: Solutions to luck and their problems
So I guess here's the way I think about it.
In warcraft 3 you are concerned because if critical strikes are random, then some games will have droughts of critical strikes or sudden bursts of many critical strikes which will throw the game, in a way which the designers feel is unacceptable to the balance. This is partly because IMO critical strike is a poorly conceived ability, but they didn't want to get rid of it.
There's only two really obvious ways it could work though. Besides just having it be random, you could also make it fully deterministic. Like, if you have a 15% chance for this, you could have a counter associated to the unit. It starts at 0, each time you add .15. Each time it gets above 1, then you subtract 1, and award a critical strike. This will ensure you get critical strikes at pretty much exactly the correct rate.
However it's obviously not so good because if the game is played competitively, players will try to count the strikes... and it will just make the game much more annoying to play. We don't really want players to try to memorize the past, we want them really to play as though it were random, and just passively enjoy the benefits of having less variance.
To get the benefits of both, what they do is this sort of hybrid. You don't quite have a counter, but you remember the past... not the distant past though. And there's some randomness still, but not as much. So it's some sort of interpolation between the two extremes.
How do you exactly pick the interpolating strategy? Idk its a judgment call, you have to make it up to some degree. But if it causes actual players to adopt a new strategy then you have to go back and tone it down somehow...
---
In wesnoth it doesn't really make sense to have a counter like this for each unit, and it wouldn't even really prevent luck swings. The thing that is most devastating in wesnoth is when, on a single turn, all or most of your units hit a lot, and all of the other guy's units miss alot. So if all of the individual units are decoupled it won't really solve the problem.
Besides this it's more complicated because each unit can have many different cth values in a game, so it's not even clear how the fully deterministic one works. I guess there are various no luck modes that that try various solutions... but some specials may get sort of corrupted in the process, like slow for instance.
As I understand, one kind of low discrepancy sequence (I guess there are many possibilities) that would work in two dimensions (for the unit square) is to just iterate through all the points of a grid, of increasingly finer and finer resolution. (The resolution gets arbitrarily small, so that the sequence is technically infinite) The point of this is that you can sample any number of points that you like, and for any smaller square, the fraction of points that hit that square is going to be close to the expectation -- the error is going to depend on the number of points you picked, and it will actually be smaller than if you had chosen that same number random points.
So if you used such a sequence for attack values, it will mean that about the right number of 60% events hit, about the right number of 70% events hit, etc., without you having to track all of these possibilities separately with a bunch of counters for each possibility and each unit or something. (You might want to have a different sequence for each team though, as that's what will enforce that no team was significantly luckier than another.)
Because you want it to be somewhat random, you can use this "random shift" idea, so that if you might choose a random shift number, say .1557, and then in the game 60% events will correspond to the window [.1557, .7557], and 70% to [.1557, .8557], etc., relative to the low discrepancy sequence. So that will somewhat randomize it, while still giving you a guarantee on the rate of 60%, 70% attacks etc.
However as you also point out, the whole thing might corrupt the player's strategy by allowing them to adopt Gambler's fallacy. So as in the WC3 idea, you probably want to do some kind of interpolation between the two extremes.
The most obvious way to me to do that would be to periodically reseed the random shift number, say after 100 attacks or something. The frequency of reseeding is your interpolating parameter - if you reseed at every attack, then it's exactly the same as if there was no low discrepancy sequence and it was fully random. If the parameter is infinitely long, then it's like you are only using the low discrepancy sequence. So you'd might have to play with it a bit and see what feels natural / what leads to players taking corrupted strategies.
You might also want to restart the low discrepancy sequence also when you reseed, that's a more subtle detail, I'm not really sure... you would have to think about what it means in terms of the low discrepancy guarantee. You might also want the periodic reseeding to be not a fixed number but a random number, so like, at every turn there is a fixed chance to reseed and restart.
Anyways its just an idea, it would take some thought and some tinkering to really flesh it out.
In warcraft 3 you are concerned because if critical strikes are random, then some games will have droughts of critical strikes or sudden bursts of many critical strikes which will throw the game, in a way which the designers feel is unacceptable to the balance. This is partly because IMO critical strike is a poorly conceived ability, but they didn't want to get rid of it.
There's only two really obvious ways it could work though. Besides just having it be random, you could also make it fully deterministic. Like, if you have a 15% chance for this, you could have a counter associated to the unit. It starts at 0, each time you add .15. Each time it gets above 1, then you subtract 1, and award a critical strike. This will ensure you get critical strikes at pretty much exactly the correct rate.
However it's obviously not so good because if the game is played competitively, players will try to count the strikes... and it will just make the game much more annoying to play. We don't really want players to try to memorize the past, we want them really to play as though it were random, and just passively enjoy the benefits of having less variance.
To get the benefits of both, what they do is this sort of hybrid. You don't quite have a counter, but you remember the past... not the distant past though. And there's some randomness still, but not as much. So it's some sort of interpolation between the two extremes.
How do you exactly pick the interpolating strategy? Idk its a judgment call, you have to make it up to some degree. But if it causes actual players to adopt a new strategy then you have to go back and tone it down somehow...
---
In wesnoth it doesn't really make sense to have a counter like this for each unit, and it wouldn't even really prevent luck swings. The thing that is most devastating in wesnoth is when, on a single turn, all or most of your units hit a lot, and all of the other guy's units miss alot. So if all of the individual units are decoupled it won't really solve the problem.
Besides this it's more complicated because each unit can have many different cth values in a game, so it's not even clear how the fully deterministic one works. I guess there are various no luck modes that that try various solutions... but some specials may get sort of corrupted in the process, like slow for instance.
As I understand, one kind of low discrepancy sequence (I guess there are many possibilities) that would work in two dimensions (for the unit square) is to just iterate through all the points of a grid, of increasingly finer and finer resolution. (The resolution gets arbitrarily small, so that the sequence is technically infinite) The point of this is that you can sample any number of points that you like, and for any smaller square, the fraction of points that hit that square is going to be close to the expectation -- the error is going to depend on the number of points you picked, and it will actually be smaller than if you had chosen that same number random points.
So if you used such a sequence for attack values, it will mean that about the right number of 60% events hit, about the right number of 70% events hit, etc., without you having to track all of these possibilities separately with a bunch of counters for each possibility and each unit or something. (You might want to have a different sequence for each team though, as that's what will enforce that no team was significantly luckier than another.)
Because you want it to be somewhat random, you can use this "random shift" idea, so that if you might choose a random shift number, say .1557, and then in the game 60% events will correspond to the window [.1557, .7557], and 70% to [.1557, .8557], etc., relative to the low discrepancy sequence. So that will somewhat randomize it, while still giving you a guarantee on the rate of 60%, 70% attacks etc.
However as you also point out, the whole thing might corrupt the player's strategy by allowing them to adopt Gambler's fallacy. So as in the WC3 idea, you probably want to do some kind of interpolation between the two extremes.
The most obvious way to me to do that would be to periodically reseed the random shift number, say after 100 attacks or something. The frequency of reseeding is your interpolating parameter - if you reseed at every attack, then it's exactly the same as if there was no low discrepancy sequence and it was fully random. If the parameter is infinitely long, then it's like you are only using the low discrepancy sequence. So you'd might have to play with it a bit and see what feels natural / what leads to players taking corrupted strategies.
You might also want to restart the low discrepancy sequence also when you reseed, that's a more subtle detail, I'm not really sure... you would have to think about what it means in terms of the low discrepancy guarantee. You might also want the periodic reseeding to be not a fixed number but a random number, so like, at every turn there is a fixed chance to reseed and restart.
Anyways its just an idea, it would take some thought and some tinkering to really flesh it out.
-
Sanae_Kotiya
- Posts: 9
- Joined: June 1st, 2013, 6:59 am
Re: Solutions to luck and their problems
Why not just update to a new RNG system that everyone can agree is fair?
Re: Solutions to luck and their problems
If by "fair" you mean that Wesnoth's RNG is not producing sufficiently randomized numbers, then you have entirely missed the point of this thread. If by "fair" you mean something else, then please clarify.Sanae_Kotiya wrote:Why not just update to a new RNG system that everyone can agree is fair?
http://www.wesnoth.org/wiki/User:Sapient... "Looks like your skills saved us again. Uh, well at least, they saved Soarin's apple pie."
-
ChaosRider
- Posts: 846
- Joined: April 15th, 2012, 1:15 pm
Re: Solutions to luck and their problems
Well, i allways believe that ai sides have somehow increased chances for each hit (because they dont have much missed strikes...) that's why i prefer units with high hp and res than units with high terrain defense (which allways s*cks vs ai army cause they dont care about your terrain defense) and low hp & res. From a long time i think to do some code which will decrease ai chance to hit (atleast for -10%) because as i said ai units were much too lucky. Ofcourse even with ai op luck you can beat him (its require some practise and find own way to do this) but still its really irrating when you end your turn with -50% done damage and +50% taken damage...
When we talk about chances to hit it could be nice if high lvl assasins (from 3 lvls) had some epic dodge (but i guess this part of my post require own topic), for one of each 3 successful hits they could have 50% chance to dont get hit.
When we talk about chances to hit it could be nice if high lvl assasins (from 3 lvls) had some epic dodge (but i guess this part of my post require own topic), for one of each 3 successful hits they could have 50% chance to dont get hit.
Creator of WOTG (+2880 units), MWC (+615 units), SurvivorsArea, RandomColosseum, RC WOTG, RC MWC, ColosseumRandomClonesBattle, BetweenDarknessAndLight, StealingWeapons, MoreUnitsForms, MoreDamageTypes, CanBeOnlyOne, ColosseumOneWinner, BonusSpam, CriticalStrike - available at 1.12 Wesnoth server.
Re: Solutions to luck and their problems
Not in mainline.ChaosRider wrote:Well, i allways believe that ai sides have somehow increased chances for each hit
Author of the unofficial UtBS sequels Invasion from the Unknown and After the Storm.
Re: Solutions to luck and their problems
Not that I'm important, but I like the idea of a slow and steady implementation of solution #5.Suira wrote: Solution 5: extra abilities/traits that influence luck
This solution also somewhat solves something I also personally consider to be a problem: namely that units are often too similar. The idea is to add abilities and/or weapon abilities that interact with luck.
Examples:
- Ability "nimble", can never be hit 3 times in a row while defending. Balance with -1/2 hp.
- Weapon ability "feint", can never miss 3 attacks in a row while attacking or hit 3 attacks in a row while attacking.
- Ability "disengage", if you miss all your attacks while attacking, regain 2 movement points. Balance with -1 damage. Alternatively, does not consume leftover movement points when all your attacks miss.
- Weapon ability "reach", deal 1 damage each time you miss. Balance with -1 damage.
- Ability "frustration", each time you miss, your next attack is 10% more likely to hit, each time you hit, your next attack is 10% more likely to miss. Does not compound and is reset after every fight.
- ABility "guardian", while defending, the first time you are attacked by an attack that could kill you, you are certain to evade that attack (it could have missed without this ability, but now you are sure). Can activate once a day, i.e. every 6 turns. Balance with -5 hp.
- Weapon ability "finisher", while attacking, if your last attack could finish the enemy, it is twice as likely to hit. Only given to weapons with 4+ attacks (preferably 5), balanced with -1 damage.
- etc, etc...
Obviously not all units would have traits like this, but rather you'd give it to units that are too similar to other units to differentiate and/or promotions that are underused. Like poacher, javelineer, elven fighter, lancer, elvish sorceress, etc. Optimally each race should have some abilities like this. Note that statistically these abilites are all pretty weak, especially the first 2.
Re: Solutions to luck and their problems
Solution 5 sounds interesting, and could even be included easily? I would just be happy if a unit with, say, 2 or more strikes, were set to miss all of them because of the RNG would actually hit on the last strike with half damage, rounded up. That could very possibly reduce the frustration levels a ton.
Edit: instead of the last hit inflict 50% all the time the damage should be determined by the chance to hit, so if a mage misses the last hit will do 70% of normal damage. Yeah I like that
Edit: instead of the last hit inflict 50% all the time the damage should be determined by the chance to hit, so if a mage misses the last hit will do 70% of normal damage. Yeah I like that