Originally Posted by
Rustjive
Symer - I don't really understand how you can write so many words on the merits of not Hit capping when your argument is based entirely off of a SimCraft profile that implicitly devalues hit. You don't have any data to stand on, really, to be making arguments that use so many significant digits.
Are you sure I have an argument? Or do I have one that would require anything other than the default simc profile? Usually I just point out that this profile is doing everything you think is missing from actual playing. Or I talk to you about what the simc results are saying, and how they happen. Some things that do not happen in the default list are maybe coming up in your imagination, but I should be the one who gets to demand you to prove that they are significant in quantity and size.
Can you base the opinion that crit rating is the worst stat for affli on anything but the simc?
I thought I was using less significant digits than what Bonkura did, when we even spoke of the same thing. I know it was probably accurate what he said, but removing the digits made it look better for me. Sorry for trying something so cheap. Last time I used 3 or more digits was also five weeks ago, if you could please quote me to make it easier for the poorer readers. And make yourself an error analysis before you speak of a result in this place anymore, really.
I write to you here about the accuracy of things mentioned later in this post.
At the bottom of the post there is a rational number 1/total_ticks. This requires inserting more data from the chosen dot and haste and I get them from game or from online if I have to. 1/total_ticks came into being however when one assumes that time is divided into periods that have a length of a tick. Values of 1/total_ticks are still infinetly accurate and tick does not have to be defined. Proc is thought to occur at any of theese ticks with equal probability. This is an approximation where you are not having (internal) cooldowns, and no dots that always sync with cooldowns, but rather a chaos between them, which is more natural to real game than what it is for simc. Yet it will end up chaos in a long fight and you can pull several worldoflogs or simc logs and check how much one proc with icd can be delayed at the beginning of a fight and mimic it at the next proc etc. Picking maybe samples would also make an distribution for a proc start time. If multiple delays can easily get wider than the full duration of the dot before even a simple fight gets to the middle of its duration, then my approximation is pretty good.
Time remaining on dot when proc runs out is distributed so that lowest value is total_duration/2 (because under that it would have gotten refreshed before proc runs out) and highest is 2total_duration/3. Mean value is total_duration. When time is a discreet variable
I compare the approximation of flat distribution to one in-game log, by looking at last refresh event before a proc runs out. So I am looking at Corruption and Inner Brilliance on the world 3rd affliction dps run on Ta'yak, where the person has only Corr and SB:swaps dealing with the dot:
http://www.worldoflogs.com/reports/r...?s=3610&e=4008
Resulting seconds are such that if they approach 0 in the log they correspond with 3total_duration/2 in the theoretical distribution, while <18 is the left side of the approximation. Any number higher than 18 would mean one more doable cast. Forms of the distributions between their end points are expected to be same.
with this query
Code:
[{"spellNames": ["Corruption"], "eventTypes": [6], "actorNames": ["Marilina"]}, {"spellNames": ["Soul Swap"], "eventTypes": [6], "actorNames": ["Marilina"]}, {"spellNames": ["Inner Brilliance"], "actorNames": ["Marilina"]}]
there is
Code:
[19:51:38.085] Marilina casts Soul Swap on Blade Lord Ta'yak
[19:51:50.238] Marilina's Inner Brilliance fades from Marilina
12.153
[19:52:26.797] Marilina casts Corruption on Blade Lord Ta'yak
[19:52:35.333] Marilina's Inner Brilliance fades from Marilina
14.506
[19:53:20.827] Marilina's Inner Brilliance fades from Marilina
[19:53:23.058] Marilina casts Corruption on Blade Lord Ta'yak
2.231
[19:54:07.795] Marilina casts Soul Swap on Blade Lord Ta'yak
[19:54:09.161] Marilina's Inner Brilliance fades from Marilina
1.366
[19:54:45.650] Marilina casts Corruption on Blade Lord Ta'yak
[19:54:54.265] Marilina's Inner Brilliance fades from Marilina
8.615
[19:55:27.847] Marilina casts Soul Swap on Blade Lord Ta'yak
[19:55:45.842] Marilina's Inner Brilliance fades from Marilina
17.995
[19:56:09.434] Marilina casts Soul Swap on Blade Lord Ta'yak
[19:56:10.946] Marilina gains Inner Brilliance from Marilina
[19:56:30.963] Marilina's Inner Brilliance fades from Marilina
21.529 -> 3.529
[19:57:06.451] Marilina casts Soul Swap on Blade Lord Ta'yak
[19:57:16.896] Marilina's Inner Brilliance fades from Marilina
10.445
[19:57:51.818] Marilina casts Soul Swap on Blade Lord Ta'yak
[19:58:03.738] Marilina's Inner Brilliance fades from Marilina
11.920
m=9.1956
\sum_{i=1}^{9}(x_i-m)^2 /9 = 265.615
s^2 = 29.5128
s=5.4326
So the time of refresh before proc runs out has a mean close to total_duration/2, and it could have a std. dev of 5.435 which makes a quite flat distribution.
Then also there is a number below which is called variance, picked from a site, where we have std. dev given with 8 significant numbers, as is the avg dps. Error in the dps has a reported ratio of 4*10^-4 so one could go with only five, and basicly consider the dps error of a single run as absolute error of 57.90 ~= 60. If std.dev is proportional to a number of simulations ran with this error, it will have the same error ratio. However the variance is proportional to squared variable(s), which have rational error same as the dps has, and therefore the rational error is then considered twice as large: 8*10^-4. So when I get 8727565.4 for the variance it would have absolute error of ~3000 when rounding up and the value I am allowed to speak of is 8 727 000 +/-3 000.
Originally Posted by
rezoacken
Originally Posted by
Symer
You mean that the only way to compare two profiles is to put two people together and let each one throw a dart into their own distribution.
Yes, repeated infinitely. If you want to know when one is better than the other (as a %) I wrote how you do it. Of course non hit cap will have a better mean on average. It just doesn't mean it's a guaranteed win.
A dart. Killing the distributions is not necessary for anyone. Once they were generated with enough sims, they now merely manifest to you so that one player wins in a single throwing match at a time.
Originally Posted by
rezoacken
Originally Posted by
Symer
With the picture and the interpretation I assume that you have one person to randomly pick one percentile. It can be choosen from the hit cap distribution, where randomnes had to do with crits, procs etc.. When you change one part of your profile, you aren't necessarily affecting any part of your luck with those things.
What does it even mean ? If you change your % of something happening you change your luck. Or a better value for luck is called variance (standard deviation). The closer something random is to 50% chance of happening, the bigger the variance (hence the bigger your "luck"). Hence, if you increase crit from 20% to 25% your variance grows. Changing stats affect luck. Going from 0% miss to 1% miss increases variance.
Nevermind what you call variance. Way in which all iterations are increasing their dps happens like follows.
There is a one spell fight, where one spell can land. That spell does 100 damage, crits for 200 at some rate such as 50 %, and has no other randomness.
That is profile A
Now even if you add spell power, so that for Profile B the spell starts hitting with 110/220, you get to the situationg where Profile B only wins 50 % of the time. Yet it is clear that getting more spell power is increasing dps 100 % of the time, because after each individual fight you can say that when Profile A did 100, he would have done 110 with the Profile B settings. And you don't have to think that maybe the crit chance rerolls again. Therefore individual iterations between theese two profiles are almost linked together.
To me, adding 2-3k mastery also increases damage 100 % of the time. And maybe I should have drawn you 4 distributions where you add the mastery first and then remove hit, and then look at removing hit from the first setting perhaps. But it will only look like you move distributions to left and right all the time. Then I just happen to think that when high mastery profile gets the hit removed but stays ahead of the first one, it might just be similar to increasing dps of individual iterations as in the example, or as in the first step. Hits taken away were only a hindrance on the way of the mastery boost.
Originally Posted by
rezoacken
Originally Posted by
Symer
When doing this with hit % I almost assume that the number of misses is always average, and when you put average amount of misses to any iteration of a hit cap profile the dps increases 95 % of the time. Though it would be 100 % this way, and random amount of misses are already affecting the numbers.
So first you pick a random percentile and then you pick the average and since one average is better than the other you conclude: the dps increases 95 % of the time. Non-sense as proven before.
I am not doing anything with the percentile in your sentence. You also say that I said something about avg. dps, while I do not mention such a thing in previous quotes. But yes one can asume alot if one average dps is higher than another. And it does good if you first ponder what concepts of "increases" and "of the time" are meant with, by others than yourself.
Again I say that the individual fight that is in percentile P for hit cap, is same as the one in P for non cap. Since for all P > 5, dps_no_cap > dps_cap, then basicly non-capping increases dps more than 95 % of the time. It makes more sense in game than it makes in notebook math, but you can't please everybody.
Originally Posted by
rezoacken
Yeah because that's the cold hard mathematical truth. A small increase in dps will not be apparent most of the time.
Does it mean you should just shove it away ? No, after all a big difference is the sum of many small difference in WoW.
But if it means you would have to change your playstyle, it is stupid to affirm that's it's something everyone should do.
Do you get all the gear in your guild by saying this stuff?
Originally Posted by
Bonkura
Originally Posted by
rezoacken
Originally Posted by
Symer
And when some professional is using procs, even without having lower min. dps he is still just gambling because chance to get more dps is really only ~50 %.
Your post is barely comprehensible and I don't think you know what you're talking about but I'm bored.
Incomprehensible.
Thanks. I didn't even know where to start after his first post in this thread. I'm glad I wasn't the only one feeling this.
No one has said anything about my first post, ie. I see that it has not been "dismissed" after one month of studying. You should not become prejudiced (... chronologically afterwards) but still keep asking yourself, how do you know now that there is nothing wrong in your head when you read that post?
I think Bonkura is still bluffing, about the simc action list he seems to be refering to from time to time. He still can't have it, because he is so smart he would have seen that when he runs it, it produces dps distributions.
Distributions with proc abuse are of same type as in my picture, so they mostly overlap with the default ones, as they have avg. dps that is ~1 % higher. You can print them and throw darts into them like you are doing for the others already.
To rezoacken:
"Professional" refers to another profile which is simc-like, but not your simc. "Procs" are things that give better stats momentarily, instead of permanently like reforging does, but they are still some good for dps. If you read my whole post, with the picture, there was a speculation towards someone else there, about having minimum dps dropping when using procs in game. However it was set aside in my sentence and you don't need to know so badly about what I set aside. This "gambling" is like tossing a coin, with a marginal chance of winning close to 50 %. Reforging was also gambling, but I added a thing called variety to expressing all the features of my examples, the way they mean it in literature. "Getting more dps" is same as it was in your version of who gets to win, and as it is in the previous example. So it means "Professional" is throwing a dart on the right side of the guy, who's name is proc-free default profile, while both of them are landing one dart randomly inside their respective distributions.
Originally Posted by
voidspark
2) Suffice it to say I don't think hit cap versus not hit capping is a "Simcraft" matter. There's no debate. Yes, "not hit capping" "might" get you a marginal (less than the variance over 20 attempts) amount of DPS. Yes, Simcraft is a computer and reacts with lightning speed always doing exactly what the priority order, no matter how complicated ("miss_react") says.
We can debate if it is a simcraft matter. I claim that nothing in game is different enough from simcraft. And you need to say what is. But you are also a hypocrite because simc is the only thing you know, when you state something about what we are talking about in the possible gains, and you want them to prove your point, as if they were low. It is like a pessimist opinion, but not quite, because you would welcome that kind of gain from anything else.
Also the picture I made is clearly meant for you. First though in http://www.simulationcraft.org/510/Warlock.html affli results, variance is 8.727 +/-0.003 million dps, so I think you made the record. However when we talk about statistics that require maybe 20k samples to be accurate, don't calculate this measure from 20.
But what you were maybe thinking was that dps gains of the size of the increase in std. dev. are not nice?
If it was so then assume every real fight is actually resulting in the avg. dps +/- std. dev area. Edges of distribution are cut off and same amounts of results are inside what remains.
Define
capped avg. dps: m
capped std. dev: s
non capped avg. dps: M
non capped std. dev: S
M-m = S-s
Bottom dps results are the same for both profiles, because m-s = M-S.
Close to their means, values of non cap are higher until M > m.
Then at the top dps M+S > m + s
At the edges, non capped profile has more weight than the capped. But since distributions f, F at this interval are mirror symmetric over their respective avg. dps, or f(m-s)=f(m+s) etc., it means that 50 % of selected results in hit capped distribution are under m. Non capped distribution has its 50 % of results above M, and also results between [m, M] are above m. Theese are resulting as more victories in darts.
More results above m means non cap profile also has less results below m, meaning it does have a smaller chance of "getting lower than average results", where that average is m from the hit cap profile.
In simc, or my picture, the case is that accurately rather M-3S=m-3s than M-S=m-s.
Originally Posted by
voidspark
Yes, Simcraft is a computer and reacts with lightning speed always doing exactly what the priority order, no matter how complicated ("miss_react") says.
Why don't you do the repair on react_time yourself, and simulate with different values, if you think it makes all the difference?
Originally Posted by
voidspark
But most top end raiders have said that in fact, having to react to a miss is worse than just hit capping and taking the marginal (less than the variance over 20 attempts) DPS loss, for more reliability on both encounter mechanics (when they are hard) AND the DPS checks.
Those people lie. They do not know if their their react time is too big to loose the benefits from not hit capping, unless they measure it with a stopwatch, and unless you do what I asked.
What knowledge allows a top end raider, to tell what significance their (~1k avg.) gear decisions have on a fight (where dps can vary ten thousand due to everything) when, because of your ideas and standards, there is not a single model/method/program, or clearly a brain, that could analyze and transmit the information of the real fight?
Hard is not enough, if you even like this game. But dps-amount-wise every dps check becomes easier to meet, if you make decisions that increase dps 95 % of the time or there abouts.
Originally Posted by
Bonkura
I've played without hit for most of my progress this tier. The RNG wasn't that bad for me but it's hard to prove it and thus really hard to justify it as well. Much of what we're discussing here is tough to put in perspective to form the theorycrafting being done because we're adding many more factors that are not easy to put on paper. However a conclusion that shouldn't be that difficult to make out of it is that the standard deviation from not being hit capped will greatly increase when we add these factors. This goes beyond a SimCraft APL that reacts to procs.
Also I'm sorry about the statement regarding ~1k DPS gain from non-hit cap in full T14H BiS. It's probably less with the APL we are using but that's hardly in favor of an argument based on going for very low hit rating.
I can make this argument for not capping: Bonkura is witholding evidence that would put not capping into even higher place. Bonkura is like any other hit capper, and they can not be trusted in normal conversation.
Nothing you say is beyond proc abuse list until you have shown there even is one in your usage. Because people can really suck in writing those and you could use someone else's opinion, to see if you made it too much in favor of one gear profile etc.
Now you can predict how std. devs. rise, when you had trouble knowing if dpses or stats change?
Ideas that are far fetched or of very low damage changes are not changing the std. dev. so much. Way it has changed so far from cap profile is that you suddenly have runs with +/-20 MG ticks around average, and a +/-1 million or smaller size Haunt events. Theese are based on miss chance % which is 3-8. Next there are some new (read previous stuff though) happenings with their own frequency, like a bad proc/dot sync has chance of 1/total_ticks, if the proc can come at any time. You multiply this with miss chance to have an event with the damage penalty. Also MG is clicked more often per minute than dots.
Again, when you add a factor that increases non cap std. dev., you will also increase the hit capped std. dev.. And if you wanted to increase your std. dev because it increases dps, you will also increase the avg. dps of non cap more than you do for capped. You are not ending up with anything different yet.