1. Hello and welcome to the brand new home for PlayDota!
    Please read through our Welcome thread to see what's new!
    Dismiss Notice

MKB true strike is only 63% and not 75%?

Discussion in 'Advanced Mechanics' started by mapdesigner, Dec 14, 2017.

  1. mapdesigner

    mapdesigner Member

    2,924
    148
    63
    Aug 11, 2011
    Hello,

    I just checked this link, and according to the data in the table, the actual probability is 63.2%

    so now I am wondering, are the tooltips correct or the table when it comes to chances higher than 50%?

    is vanguard block chance 50% or 45.7%? etc

    anyway, if the table was correct then it means its daedlus is better than mkb vs butterfly (because mkb improve less than 40%)

    thanks!

    Link: https://dota2.gamepedia.com/Random_distribution#Legacy_data
     
  2. EebstertheGreat

    EebstertheGreat Forum Manager Staff Member

    10,787
    88
    48
    Sep 3, 2009
    Yes, Valve deliberately uses the same inaccurate C-values that Blizzard used for WC3 (or nearly the same; it's unclear if they took the exact values from the code, reported on the old forum, or the older values obtained by experiment, which were very close to the exact ones). The exceptions are numbers which are not multiples of 5%, such as Counter Helix, which IceFrog handcoded using (approximately) mathematically correct C-values. In practice, the difference is irrelevant, since probabilities 25% and below are extremely close to exact anyway. Only for large probabilities is there a major distinction.

    Blizzard seems to have calculated the first few probabilities correctly using Markov chains, then simply fit the rest to some function like a sigmoid. Why they didn't simply calculate them all is not clear, since it isn't very difficult to do. However, in WC3, PRD is never used for values higher than 25% anyway. This is therefore only an issue in custom maps like DotA and of course games based on them like Dota 2.

    So the actual probability (averaged over many attacks) for MKB to proc is indeed about 63.2%, not 75%, meaning if we ignore the true strike component, the weapon offers an average of only 37.92 damage per attack rather than the nominal 45. (Of course, it also passively gives 60 IAS). This is still a bit more damage than Butterfly for strength and intelligence heroes and at much lower cost, with the added benefit of true strike on the procs, but with no evasion or armor. Which weapon is better is clearly dependent on the current state of the game. MKB is a bit stronger offensively than Butterfly, but offers no defensive tools. For agility heroes, Butterfly is usually better offensively as well, though not when facing significant miss chances. Daedalus is much better for heroes with high attack speed (again, unless facing significant miss chances), but not as good for heroes with low attack speed, as they will benefit more from the IAS even though they are losing a ton of potential damage. And again, Daedalus costs more than MKB for a reason.
     
    dR likes this.
  3. mapdesigner

    mapdesigner Member

    2,924
    148
    63
    Aug 11, 2011
    @EebstertheGreat
    If you dont know what values are used, how would you know the actual chance is "indeed 63.2%"
     
  4. mapdesigner

    mapdesigner Member

    2,924
    148
    63
    Aug 11, 2011
    well I managed to get 2 successive attacks w/o proccs and that means C value is certainly less than 50% and therefore the total probability cannot be more than 66.7%,

    I am still curios to find out the actual probability though, we will see

    Edit: in fact I could manage to get 3 failed strikes in a row which leads me to think it doesnt even use WC3 method. maybe it uses the PRD inverse method.
     
    Last edited: Dec 14, 2017
  5. mapdesigner

    mapdesigner Member

    2,924
    148
    63
    Aug 11, 2011
    Alright, so I have so far tested for 220 hits and I had 75.22%

    so i would believe it uses fairly accurate values.

    Thanks!
     
    Blamagenkind likes this.
  6. Blamagenkind

    Blamagenkind Member

    627
    88
    28
    Jun 10, 2009
    I don't think that's not how it works, if you are lucky you can proc MKB on every hit as long as it has a chance to proc. PRD just reduces the chance of unlucky incidents where it doesn't proc at all. Maybe that's how they made it, but I doubt it.

    Either way, we won't know the actual values until someone looks into the game files, which is probably not easy to do, otherwise people would've done it already.

    Edit: If it still arrives at ~75% in any number of hits, I'm guessing the shown values are the average percentage of procs. Which maybe the actual value 63.2% was chosen so that the average hits would be 75%.
     
    Last edited: Dec 15, 2017
  7. mapdesigner

    mapdesigner Member

    2,924
    148
    63
    Aug 11, 2011
    What do u mean by actual value. You meanthe starting C value?
     
  8. Blamagenkind

    Blamagenkind Member

    627
    88
    28
    Jun 10, 2009
  9. mapdesigner

    mapdesigner Member

    2,924
    148
    63
    Aug 11, 2011
    In prd there is no actual prob. For all hits. There is an average and there is starting C (or maybe valve developed another psuedo method)
     
  10. Blamagenkind

    Blamagenkind Member

    627
    88
    28
    Jun 10, 2009
    There is. :)
     
    Last edited: Dec 15, 2017
  11. mapdesigner

    mapdesigner Member

    2,924
    148
    63
    Aug 11, 2011
    What it means? Its the probability of what exactly? Can you define?
     
  12. EebstertheGreat

    EebstertheGreat Forum Manager Staff Member

    10,787
    88
    48
    Sep 3, 2009
    In that case the Dota 2 Wiki should be updated. Clearly my info is out of date.

    This is from DotA 1 values derived both experimentally and theoretically from disassembled code.

    PRD works by using a given starting probability C on the first roll, increasing the probability by C each time the roll fails to proc, or resetting it to C if it does proc. The nominal probability is (or should be) the limit of the average probability over n rolls as n goes to infinity; in other words, it is the long-term effective probability. If after over 200 trials, mapdesigner is still proccing 75% of the time overall, that is strong evidence that the effective probability is equal to the nominal 75%, not the 63.2% that WC3 gets due to using a highly inaccurate value for C, which was initially ported directly to Dota 2 when it first added PRD to avoid unintended balance changes.
     
  13. mapdesigner

    mapdesigner Member

    2,924
    148
    63
    Aug 11, 2011
    Thanks EebsterTheGreat

    anyway, my first tests were 10 by 10 tests, so I did 10 hits and then counted how many were procss or not, and then did 10 more hits, etc until I reached 220

    but I had doubts, as I always do, becuz maybe stopping the attack had some effect or so,

    anyway I did another test and got 75.45% chance (about 80 hits) with similar test,

    but I also did a countinuous 40 hits test with chance of 74% of procc (only 13 were not proccs).

    I will prolly add this to the DotA Lab sometime when I have time.

    Thanks!
     
  14. Blamagenkind

    Blamagenkind Member

    627
    88
    28
    Jun 10, 2009
    That's what I'm saying. What's shown in the tooltips are the average, not the actual probability.
     
  15. EebstertheGreat

    EebstertheGreat Forum Manager Staff Member

    10,787
    88
    48
    Sep 3, 2009
    Define "actual probability" in this context. Do you mean the probability of the next hit? That changes every attack.
     
  16. Blamagenkind

    Blamagenkind Member

    627
    88
    28
    Jun 10, 2009
    It changes but it is still the actual.
     
  17. EebstertheGreat

    EebstertheGreat Forum Manager Staff Member

    10,787
    88
    48
    Sep 3, 2009
    The average probability is the actual probability that a randomly chosen attack will proc, which is far more meaningful than the probability some particular, arbitrarily chosen hit will proc. I honestly don't know what you're getting at.