Suggestion Why we don't like CV

Discussion in 'General Archive' started by BigPapa, Nov 18, 2015.

Dear forum reader,

if you’d like to actively participate on the forum by joining discussions or starting your own threads or topics, please log into the game first. If you do not have a game account, you will need to register for one. We look forward to your next visit! CLICK HERE
Thread Status:
Not open for further replies.
  1. BigPapa

    BigPapa Forum Overlooker

    [​IMG]
    [​IMG]
     
    That0n3guy, misterbean and _Baragain_ like this.
  2. _Baragain_

    _Baragain_ Living Forum Legend

    That is actually very easy to explain... Well sort of easy to explain.
    Right now there are two threads with this general theme. This one, started by BigPappa, looks at the "feel" of the system. Then there is mine, which looks at the DSO team's poor technical implementation of the system.

    Here is an excerpt from mine that ought to explain what is going on with your example.
    Another issue that I did not address in my original thread (but now I may need to go back and look at adding) is the fact that a 2H weapon takes up both slots, so it is to be expected that a LH item's and a RH item's individual values would be lower than a 2H. However, with 2 good 1H items, you may have more CV than with a 2H. This isn't to say that 1H is better in all cases... we already know the CV system is flawed, it simply highlights the need to have 2 separate values, offence and defense, that are based on objective quantities as opposed to the developer's subjective opinions.
     
    BigPapa likes this.
  3. sebastian_fl

    sebastian_fl Count Count

    That is the only possible option. And there is no mathematical measure to anything beyond that.
     
  4. _Baragain_

    _Baragain_ Living Forum Legend

    Not totally true. Here is a system that would be an actual method to compare items and builds. It just takes more complicated programming than the current CV system.
     
  5. sebastian_fl

    sebastian_fl Count Count

    okay. here you go)

    I'm working on Character Optimization tool, and I've been fairly successful in recent days. I was only looking into one scenario - what are the items I should equip to maximize
    ED = MedianDamage * (1 - Crit) + MedianDamage * Crit * CritDmg

    the screenshot is here https://board-en.drakensang.com/threads/inventory-export-to-excel.69987/#post-638862

    I do not take speed into account, for now, neither I account mana/rage.

    Even without speed, mana, defensive attributes, there is not a single deterministic algorithm to analyze your items without actually analyzing every possible selection, so N number of selections where N = #amulets * #belts * #cloaks * #rings * (#rings-1) / 2 * ... * #boots

    All because items cannot be compared irrespective to a specific build/other items. So if you want CV of the item to show how good it is for this very specific currently equipped setup - yes, it is doable.

    But if you want CV to be how much ED, DPS, EffectiveHealth or whatever it brings to a character no matter the other items, sets, essences, bonuses from the trees - it makes zero sense. How do you know what is better for an undefined build? 100dmg or 800 crit? How do you know the general answer to that? - it all depends on other items you have.

    And here is a proof. Imagine you have selected a vector of items [amulet, belt, ..., boots] that represent a character build.
    Having that, you are going through all the items you have in the locker/inventory to check if that item would bring value to your vector. If it does, you replace the item within a vector, and start again. If no more items should be replaced, you decide it to be a resulting vector.
    1. The item does not necessarily brings value alone, it might bring value together with other item or items, so the 1 level search will not help.
    2. It is possible, that such an algorithm will never end, as it could end up with a loop, thanks to relationships between items.

    So you have to go through all possible combinations. But even then there are multiple issues:
    1. Speed - speed 2.0 does not mean 2x shots per minute (bug because of preambula frames, could be addressed). It means even less, thanks to different skills with different breakpoints. For example, as a mage, would speed 4.0 make you 4 times better? would you cast 4x lighting strikes and magic missiles?
    2. Mana - is 20% more mana making you 1.2x better? - absolutely no.
    3. Armor to Resist ratio, what is the best? :)
    4. Healing. The absolute gems and rings of life are easier to count, they favor armor and resist. But you need to know how many seconds would they work so you know how much HP do they bring. The other types of healing is impossible to count, you can just offer some approximate coefficient.
    5. travel speed
    6. all those guardians that appear if low hp
    etc

    Yes, some of those are additional junk that can be ignored, but even with root things like speed and mana there are issues. So, for my tool, I decided to do the following:

    1. For every character and every speed breakpoint - count real number of shots made. Define a 'real' speed as a median number of shots using skills with different breakpoint. For instance, if mage with speed 2.00 shots 180 magic missiles per minute, or 165 fireballs per minute (assuming infinite mana), take the real speed of 2.00 break as 1.725.
    2. For every character decide what is the most effective skill combination to maximize damage per minute to a static object, and measure the role of speed and mana in it, so we get coefs for mana and speed
    3. Set DPS = (MedianDamage * (1 - Crit) + MedianDamage * Crit * CritDmg) * ManaCoef * SpeedCoef
    4. Set EffectiveHealth = (HP /(1-Armor) * ArmorCoef + HP / (1-Resistance) * (1 - ArmorCoef) ) / (1 - BlockAmout * BlockChance) * HealingCoef
    5. Set CharacterStrength = DPS * EffectiveHealth / SomeConstant

    ArmorCoef - is how much damage you expect to come from elemental sources. 1- ArmorCoef - how much from magical sources

    As for the travel speed, I decided the user to set it himself, the minimal he wants to achieve.

    Currently I have only implemented Damage option, for twohanded mages and tanks. It was original intent. But it looks like I got interested in it, so will continue. I would love to share it with you, as it is freely available including all the source code, but guess Moderators will filter that out. You would surely be able to help with formulas. Any ideas on how to improve the above are welcome)
     
    Darwarren likes this.
  6. Darwarren

    Darwarren Count Count

    Excellent analysis and progress.

    I had originally used overall strength = offensive strength + defensive strength, rather than = offensive * defensive. The differences in scale might change range and usability. The numbers need to be big enough to seem significant. Whatever you decide.
    A lot of the character stats are really indexes with hidden math. Baragain had some formulas that still might be relevant. They were buried with old forum posts, I think.
    Some martial artists convert everything to speed, power, and technique; only 3 variables keep it simple. Skill of application is beyond the scope of this, but it might point the way to best technique.
    Mana effects some skills/attacks more than others. It may need math in steps or summary. Attacks without use of mana can fill time available. Maybe you can work backwards from optimal attack/minute, if that is necessary.
    Time makes an excellent constant, turning things into physics problems.
    Model building benefits from different points of view, and from discussions. The math can be incredibly complex. We need to hear from wizards and rocket scientists. :):):)
     
  7. gemon666

    gemon666 Forum Apprentice

    now now now...I see an incredible thread here...moving from unuseful CV value to how to determine a real value of an item...
    This makes me understand there's a multitude of us working on the same problem from different point of views, we should join alltogether and develop something really useful for us all :)

    Some of my contributions to the topic:

    Comparing two items should easy, if you *do not* consider the trend in upgrades. Depending on bonuses, there may be circumstances where an item equipped gives a DPI value that increments LESS than another one, if you simulate it to be upgraded to subsequent upgrades.
    So, I'd implement the comparison (and thus, the CV to show on the item) as the following
    - consider base CV = gray items of lvl 45
    - then, if you have found say an epic amulet of lvl 50 with only increment in the DPI, the CV is calculated as the absolute DPI this item brings compared to if you'd equip a gray amulet of lvl 45.
    - the system can easily recalculate your DPI due to the fact that removing your equipped amulet brings a new DPI value (lower), that the newly found epic amulet can be used as basis to calculate the incremental DPI of that new item --> ta daaaah you have the CV value of that item
    - of course any bonus should be taken into account to calculate your new DPI "if you wore that item instead of the current equipped one", %bonuses etc.
    - this doesn't address two tricky points: 1) gray jewelries as per latest releases do not have modifiers at all (indeed you can't upgrade them...did you notice? ;)) and 2) what in case of rings?​
    - the above works if you compare apples with apples and pears with pears, i.e. just DPI increments with DPI and just EHP with EHP increments. What about mixed items? what about resistance? My contribution here is the following
    - for mixed items, experience made me identify a ratio between how much a single item increments your DPI and your EHP. Looking at numbers, I've found an empiric ratio of 1:10 for DPI:EHP to compare mixed items. Of course the value is tunable, but the idea is that
    - if you have an item that brings both +DPI and +EHP, the total increment could be = +DPI +10EHP. That should be the CV to use to compare.​
    - resistance should be considered to calculate the EHP. Also here, looking at various damages produced by mobs in PVE, I've come to the conclusion that resistance impacts EHP for 25%. I've thus personally modified the EHP formula in two fractions, 75% exactly the same + 25% that counts resistance reductions (of course do not forget that block affects also resisted damage)​

    I also have to comment more...but now I must get off my train... :p
    Gogorok
     
    Darwarren likes this.
  8. sebastian_fl

    sebastian_fl Count Count

    75 to 25 ratio between the two can be inacurate. You didn't consider the 'cost' of getting % of armor and % of resist. It is cheaper for the first portion but expensive in the end. So, with that ratio you will force more expencive armor selection at the cost of other stats, hp at the end of a day.

    And overall that approach will not work :)

    You cannot compare items irrespective to a locker. For example, you might get new 800 crit ring which wouldn't produce a lot of value together with currently equipped items. However, there is 30dmg, 40%crit item in the locker that would perfectly pair with that one. So that 800 crit item loses all comparisons 1on1, but it wins if you think about overall picture.

    This is an example of an issue of 2nd level. It could go much deeper. For instance, my tool tells me to change darkness boots to damage boots if I play with no essence instead of green, even though I'm losing 15% crit bonus. It also tells me to switch to karabosa and change number of legendaries if I go with blue essence. If I disable karabosa, it will force me to change torso and a ring.

    Again, I don't know how to explain it better, but this problem is impossible to solve mathematically without checking all possible combinations with the new item. And despite being of lower stats it might very well play with other items.

    The best you would achieve is how good is this specific item is comparing to currently equipped item for this very specific setup of currently equipped items. It isn't a lot of value for me.

    I would rather want to know how good the modifiers of the items are comparing to maximum possible stats. If max dmg is 100, and item is 75, I would like to see CV or a score, you name it, of 75%.

    Otherwise, welcome to the optimization tool to check all the possibilities) it works like a charm)

    And thanks for contribution) looks like we all use similar formulas)
     
    Last edited: Dec 10, 2015
    Darwarren likes this.
  9. _Baragain_

    _Baragain_ Living Forum Legend

    Those were some huge, good posts. I'll respond later when I get to my computer and off my phone.

    EDIT: Finally got a half hour with my comp. Whew, here we go...

    First off, excellent post. I really enjoyed your methodology and thought that went into both the tool you are creating and the problem at hand.

    Now, I want to set the record straight. I am not trying to recommend/develop the tool that you are and as such, my thoughts may or may not help. What I want and have described (above) is a true way to compare the offensive power or defensive power between two toons that is empirical in nature and is not based on the highly flawed and subjective nature of the developer's opinions. Please start by rereading what I stated above (and restated here) but this time forget about your proposed tool and consider only a CV system that truly measures the strength of a build for the purpose of comparison.

    To begin, there are two fundamental stats that can be easily quantified; DPS and EH. The element that is not so easily quantified into one or the other (listed above) could be put in an arbitrary category and have their own score. As stated, I see these all added together to provide an aggregated score. This may necessitate a subjective constant be applied to one or two of them to get them into the same order of magnitude, but, so long as the two/three scores are shown separately, this is less important because offence can still be compared to offence and defense can still be compared to defense (and misc included when you need to break a tie). You brought up the handling of speed points, and, as said, it can go either way and either should be fine as this system is supposed to be an improvement for CV (making it actually useful) as opposed to the 100% definitive answer into the question of "What is the best build?" that you are asking yourself. Finally, if you look at my
    To answer this point, I draw your attention to my final bullet in my quote. An individual item would not have an independent CV. It would only take on a CV when you add it to your build and it increases or decreases your O/D CV (offensive/defensive combat value). That is why I recommend the ∆CV to see the increase or decrease that an item would have on your O/D CVs. This ∆CV would account for set bonuses, diminishing returns, gems that might be moved over, knowledge points... etc. In the end, it is the user's responsibility to come up with their build. The CV system should simply tell them how it stacks up to the rest of the players they want to share with.
    Please see the above about ∆CV.
    Again, this is why we need separate O/D CVs. Under those circumstances, we can compare apples to apples and oranges to oranges. If you read the whole post of mine that I quoted, I talk about how in some cases, DPS is irrelevant and in others, defense couldn't matter less. I've quoted that part below.
    You hit it on the head here. I an trying to recommend a system that tells you the over all quality of your final build (offence and defense) and if a new item would improve or hurt your build based on what is presently equipped. It is the job of math savvy players like you and me to argue the details of how to determine the best build, but implementing a tool of this magnitude is unquestionably beyond the scope of the present developers (evidenced by the abysmal job they did putting the original CV system together). I know that this system may not hold much value for you, but it is only supposed to be a measure of performance, not an automated tool. It still depends on a player's intuition and experience to begin the build process and, ultimately, the decision on how to balance the O/D CV would be the player's choice. Many people have pointed out to me how I might be able to improve my DPS, but it would come at the expense of my defense, and that is something I am unwilling to give up at the moment. It is that intuition and experience that no program can ever replicate, but there is nothing to stop a well designed O/D CV system from allowing me to measure this build of mine for the purpose of comparing myself to others.

    In closing sebastian_fl, I greatly respect your work here. It is already amazing and I will do anything to help that I can. It is just that right now you and I have different goals. You want the answer to the ultimate question (42 :p), but I just want a way to measure the final product that isn't misleading or subjective.

    Lets keep up the conversation and let our geek flag fly!
     
    Last edited: Dec 10, 2015
Thread Status:
Not open for further replies.