[Rogue/Warrior] Weapon Skill Adjustment Discussion - Page 30

Apate · 08/27/07, 5:08 PM

Originally Posted by Jelu

A very simple test is to use a high mitigation on the target, say 75% (easy with evasion) and a higher than 25% crit rate on the attacker the theory states that the attacker will only ever get crits with regular attacks.

It may be helpful to be on the same page with terminology. Generally, when speaking of mitigation, we are talking about armor or block. Dodge (evasion) is considered avoidance.

I do recall some testing done with a low-level character against high level mobs (I think) where every landed hit WAS a crit, or something along those lines. Don't read too much into that though, because I can't turn up the data right now and I'm trying to finish up some work. Maybe someone recalls where that was and can link it.

Jelu · 08/27/07, 5:20 PM

Originally Posted by Apate

It may be helpful to be on the same page with terminology. Generally, when speaking of mitigation, we are talking about armor or block. Dodge (evasion) is considered avoidance.

I do recall some testing done with a low-level character against high level mobs (I think) where every landed hit WAS a crit, or something along those lines. Don't read too much into that though, because I can't turn up the data right now and I'm trying to finish up some work. Maybe someone recalls where that was and can link it.

That's not the same, a factor in crit rates is the difference in levels (defense/weapon skill) of the target and attacker. A significantly high level attacker will have an exceptionally high crit rate against a relatively low level target.

sp00n · 08/27/07, 6:01 PM

Originally Posted by Jelu

There is nothing gained by having a one roll system, especially if you're willing to accept the existence of a multi roll system is also in place. There is nothing a single roll system can provide that a multi roll one doesn't.

Yes, there is something gained. With a two-roll system you're effectively losing a part of your crit rate (which is why the one-roll system was questioned the first time anyway for special attacks).

Let's say you're a rogue with 0 hit and have a miss chance of 24% (against an even level mob) and a crit chance of 20% + dodge chance of 5%.
That makes 29% of the attacks that do not land, so it would be 20% crits out of 71% attacks that can land, which is 14.2%. So you effectively just lost 5.8% of your accumulated crit chance.

Gruntle · 08/27/07, 6:02 PM

While it is interesting to discuss thesingle-roll vs two-roll theories of white attacks it would probably be better to do that in another thread. This thread is about testing the effect of weapon skill on hit, crit, dodge and parry. The problem with trying to find all of the mechanics at once is that the theories become very hard to test.

So trying to steer the discussion back to that, would a single roll theory for white attacks solve the problems we have with low crit rates in the paladin tests? Have anyone else made tests on crit and weapon skill? The "blocked crits", will not be observed in a rogue/warrior test since they attack from the back, but will be found for the paladin test where both of them attacked from front.

• Vulajin · 08/27/07, 6:21 PM

Originally Posted by Jelu

Really, yet you're doing the same now. Believe me it's something I followed for a long time and I've never seen any "proof" that it's a single roll, only some poorly worded post by a CM on the euro boards that isn't in any way conclusive, it doesn't even have a single interpretation.

As for proof, the single roll system as described on WoWWiki implies that outcomes get pushed off the table and are therefore impossible. One of these scenarios is that regular hits get pushed off the table in favour of crits. This would mean that with enough mitigation a player would only be critically hit and never regularly hit. Admittedly it was a long time ago but I tested that theory and it's just not true. A very simple test is to use a high mitigation on the target, say 75% (easy with evasion) and a higher than 25% crit rate on the attacker the theory states that the attacker will only ever get crits with regular attacks. That doesn't happen, the attacker will have both regular and critical attacks. People have countered that evasion does something special to the attacks, resulting in multiple rolls or some such, that's simply hogwash, it just changes the dodge value temporarily. It's silly for Blizzard to develop multiple combat resolution methods when one (a multi roll system) would suffice.

The issue ultimately is that you can't prove what system is in place via testing only test to try and disprove other systems. The one roll system predicts behaviours that simply don't occur so that theory is flawed. It also makes far more sense to use a multiple roll system as it scales easily, is easy to write, maintain and wouldn't need tweaking for new abilities/talents, short of a new base mitigation (miss/dodge/parry/block) or attack (hit/crit/glance) type being added.

There is nothing gained by having a one roll system, especially if you're willing to accept the existence of a multi roll system is also in place. There is nothing a single roll system can provide that a multi roll one doesn't.

These are not the official WoW forums. It is not sufficient to simply post saying "everyone is wrong" and then leave it at that with no proof or support. If you want to contest what is a pretty long-standing and, in my opinion, well-proven theory regarding the resolution of [at least] white attacks, you're going to need to bring some proof. Otherwise, you're essentially just trolling.

Dontmindme · 08/27/07, 6:29 PM

No offense Juju, but all I'm hearing from you is your assertion as a programmer of some sort that a multi-roll system makes some sense. The problem is you seem to have nothing to really back up your claims. Personally, I can see ways of making a single-roll system work and advantages to it as well. Same goes with multi-rolls.

There has been extensive testing done on this board. We have shown linear relationships for many of the aspects concerning the "attack roll" and so far, these combined relationships indicate a likelihood of a single-roll system that encompasses most of the mechanics. Clearly, there is not a separate roll for each mechanic of dodge, block, parry, hit, crit, glancing, etc. The data just doesn't add up that way.

We have data that ranges from people with virtually 0 bonus to hit to approaching (and reaching) the hit cap among other variations. Comparing other values, there still seems to be a relatively consistent relationship to this data. If there were a two-roll system (for white attacks), where hits are determined prior to crits, this would theoretically show that those without as much to-hit will be critting noticeably less than those at the hit cap. So far, I do not see that trend in analyzing the data although it appears such a trend for yellow attacks.

I'm not going to rule out anything in analyzing the data, but so far, I believe its a one-roll system. Let me also assure you, through about 5 or more threads of posts, those on this board take "blue" posts with a significant grain of salt. We have not arrived at a one-roll system through happen chance nor undying loyalty to the power of "blue" posts.

Maybe instead of continually repeating yourself and patting yourself on the back on your belief of the 2-roll system, if you would kindly show us how what we've observed so far better fits a multi-roll model, we'd give your posts credit. But right now, all I see are more useless posts asserting a multi-role system with nothing substantial to back it up besides your so-claimed programming knowledge.

But to your claims, first, in your example of Evasion and crit chance, I know of no theory that gives boss mobs a 25% chance to crit. Right now, all of the theories deal with PvE. We know some aspects of PvP are different but are thus far unexplored. Second, could you please produce a combat log, showing both normal hits and crits in a situation where hits would theoretically get pushed off the table. If you want to disprove this, show us something.

To your thoughts an the inefficiency of a one-roll system...
Here's one way it could be implemented in a CPU friendly manner.
1) Roll once.
2) Check result to "stat 1 chance" (let's say "Miss")
3) If lower go to stat 1 outcome function (i.e. implement "Miss").
4) Subtract "stat 1 chance" from roll
5) stat counter = stat counter +1
6) Go back to step 2 for next stat

It's simple, it's relatively CPU friendly and I don't see the "more resources" that you are referring to verses a multi-roll system, outside of possibly 1 subtraction. It is still tweakable, in fact Blizzard did for the expansion by adding in the combat ratings system so that stats don't cap and push things "off the table" (except with special abilities). In fact, programmed as above, there isn't really a table, just a theoretical one that we could create on paper.

Work with us and we'll work with you. Provide us with something meaningful and tested (or an actual testing mechanism) besides just your unsupported assertions and then maybe we'll see who is right. Until then...

Jelu · 08/27/07, 6:30 PM

Originally Posted by sp00n

Yes, there is something gained. With a two-roll system you're effectively losing a part of your crit rate (which is why the one-roll system was questioned the first time anyway for special attacks).

This isn't a good test, recklessness specifically increases crit rate, what you would want is a low crit rate and high dodge to mitigate false positives (crits that would have been crits regardless of the dodge rate). Of course you don't want too much dodge as you do need some data to sample.

Let's say you're a rogue with 0 hit and have a miss chance of 24% (against an even level mob) and a crit chance of 20% + dodge chance of 5%.
That makes 29% of the attacks that do not land, so it would be 20% crits out of 71% attacks that can land, which is 14.2%. So you effectively just lost 5.8% of your accumulated crit chance.

I'd argue that's accurate behaviour, you can't expect a crit from a miss, but that's not the issue. That sort of behaviour can easily be added in to a multi roll system by simply having crit evaluated prior to hit/miss. Blocks can then evaluated after all other avoidance methods to factor in blocks on crits if that's possible (don't recall off hand).

The point I'm getting at is that a single roll system isn't nearly as robust as a multi roll one and whatever desired behaviours exist in a single rolls system can be achieved in a multi roll one while maintaining a general purpose behaviour and therefore requiring far less maintenance. A multi roll system is also easier to implement, have it's factors adjusted and requires less resources than a single roll system. In short I see no reason to prefer a single roll to a multi roll system. If anyone has any arguments as to why a single roll system might be better than a multi roll one I'd honestly like to hear them.

WibbleNZ · 08/27/07, 7:08 PM

Regardless of the underlying implementation (a multi roll system could be done as a single roll by pre-tabulating all the results); the observed behavior is better described by a system in which events push each other off the table, rather than one where they combine independently.

The one roll theory predicts that crit% is measured across all launched attacks (20% in the above example). The two roll theory predicts that crit% is measured across landed attacks (14.2% in the example). So far, evidence favors the former.

Leguaris · 08/27/07, 7:24 PM

Originally Posted by Jelu

I'd argue that's accurate behaviour, you can't expect a crit from a miss, but that's not the issue. That sort of behaviour can easily be added in to a multi roll system by simply having crit evaluated prior to hit/miss. Blocks can then evaluated after all other avoidance methods to factor in blocks on crits if that's possible (don't recall off hand).

The point I'm getting at is that a single roll system isn't nearly as robust as a multi roll one and whatever desired behaviours exist in a single rolls system can be achieved in a multi roll one while maintaining a general purpose behaviour and therefore requiring far less maintenance. A multi roll system is also easier to implement, have it's factors adjusted and requires less resources than a single roll system. In short I see no reason to prefer a single roll to a multi roll system. If anyone has any arguments as to why a single roll system might be better than a multi roll one I'd honestly like to hear them.

First of all, when crit is calculated in a multi-roll system it doesn't matter when the "Crit dice" are rolled. (% chance to hit)*(% chance to crit) = (% chance to crit)*(% chance to hit)

Second of all, this thread is not about the advatages of disadvantages of different roll systems, and theorycrafting in this thread is not philosophy, it is science. What is being done here (and there is some seriously good work in here, I wish I had three spare hours to auto attack mobs) is trying to model Blizzard's equations by collecting data, building theories and equations based on that data, and collecting more data in the hopes of proving or disproving those theories. Since I am not a programmer I don't know about the advantages of different roll systems, but I do know that one cannot go about saying how the world (of Warcraft) works by guessing by trying to psych out Blizzard. If we were still doing science by debating the philosophical merits of the issues we'd still believe the Earth is the center of the universe.

Kaubel · 08/27/07, 9:47 PM

Originally Posted by Leguaris

this thread is not about the advatages of disadvantages of different roll systems

This.

• Shalas · 08/28/07, 2:10 AM

Warrior tanking mechanics have shown quite clearly that npc white attacks are single-roll. NPC specials can't crit, and therefore must be single-roll.

Rogue specials have been shown to be two-roll. Presumably other specials are, but I've never seen any specific testing.

Fungal Bloom shows that (at least pre-TBC), white attacks behave as predicted by the one-roll system, not the two-roll system. With 80% crit, no normal hits occured (everything was a crit, glance, miss or dodge). With a two-roll system, this would only happen at 100% crit. In addition, in the tests used to show that Backstab is two-roll (with boosted miss rates), none of the autoattacks had normal hits, as expected by a one-roll system.

Autoshot was at least some point two-roll, unless we have some other explaination for blocked crits. Fairly irrelevant for hunters, though, as they tend to be at or near the hit cap.

Anything contradicting one of these statements really needs to be backed up with some form of proof, as they've all been shown quite clearly. Some other system being better in some way is not proof, unfortunatly -- if the cleanest formula was always the right one, this thread would be a lot shorter.

(Also, why specials are two-roll does have a fairly simple explaination: The game was originally designed as one-roll. At some point, they realized that adding +hit would lower Backstab's damage per energy with a one-roll system (see the whine threads about Suprise Attacks when the talent was first announced for math). As specials were already handled seperatly (as they have a different miss rate), they modified only the part that needed to be changed.)

sp00n · 08/28/07, 5:49 PM

Just one last note to the multiple rolls theory. About one year ago (actually exactly one year and 7 days) I wrote a little php script to demonstrate the effects of multiple rolls on the outcome of your attacks.
You can find it here.

Now if you insist on further discussing this issue (and who knows, if you provide us with a formula or proofs, this might actually turn out useful!) I'd like to suggest that you open up a new thread, as mentioned already here. We really should keep this thread to testings of weapon skill.

Olgas · 08/28/07, 10:45 PM

Originally Posted by Karmon

According to the modified wiki formula you need 351 hitrating to never misss with 353 skill
and 5% precision:
(24 + (365-353)*8/30 - 5)*205/13 = 350.076 (the -5 is for precision)

So I ran Letheras with 353 skill and 351 hitrating.
Out of 2542 autoattacks no misses. So far so good.
Yesterday I run Karazhan with 350 skill to see if I could get a miss.
Out of 3989 autoattack also no miss.
This is not yet enough data to refute the modified wiki therory.
350 hitrating is only 0.005% less than the expected value so I need about 20000 hits to see one miss.

Originally Posted by sp00n

As for my testings, I now have made 1162 attacks without a miss with 351 hit rating and 352 weapon skill. If it was 0.267%, I'd have seen a miss after about 480 swings, which obviously didn't happen as I have more than twice the amount of swings.

So after that I switched to 350 hit rating, and on the very first test I encountered a miss (intererestingly, with 350 hit rating I would have needed exactly 20500 attacks for a miss to occur. Talk about nice numbers, eh?).

This leaves us with the following:
(28-5-(350/(10*82/52)))/2= 0.4024
(28-5-(351/(10*82/52)))/2= 0.3707

1 point of weapon skill equals less than 0.4024% hit.
1 point of weapon skill equals more than 0.3707% hit, at least within the sample size of 1162 attacks (which is not all too much I admit).

So a 0.4% formula might at least be possible.

Hi Sp00n,

I see you and Karmon have been busy while I was away!!

I'm trying to catch up on what I've missed over the last week. I understand Karmon's post above and the conclusion he reaches. (And I'm bummed about the .2% thing not working! Really liked that formula!)

But I'm struggling a bit to understand your last test. I understand the point you made that, if it was .26667%, you should have seen 1 miss in 480 attacks. And you did double that many attacks and still didn't see any misses. OK, I understand that much. But I'm not sure that disproves the .26667% theory yet. With your number of attacks, you were likely to have observed 2 misses, and yet you saw 0 misses -- that is definitely not helpful to the .26667% theory. But, on the other hand, I don't think it is fatal yet either, since 0 misses versus 2 misses is not a huge variance, given the relatively small sample size.

And then I got confused by the conclusion you reach when you switched to 350 hit rating. At 350 hit rating and 352 weapon skill, your chance to miss would be .2715% (i.e., under the .26667% theory). So, I think you should see a miss roughly every 368 attacks. Yet you suggest you should only see a miss every 20500 attacks, which is where I think I got lost a bit...

So, ultimately you reach the conclusion that weapon skill must equal at least .3707% -- which is the part I'm struggling with.

Sorry, if I've misread something. But I'm confused.

sp00n · 08/29/07, 4:38 AM

Originally Posted by Olgas

Hi Sp00n,

I see you and Karmon have been busy while I was away!!

I'm trying to catch up on what I've missed over the last week. I understand Karmon's post above and the conclusion he reaches. (And I'm bummed about the .2% thing not working! Really liked that formula!)

But I'm struggling a bit to understand your last test. I understand the point you made that, if it was .26667%, you should have seen 1 miss in 480 attacks. And you did double that many attacks and still didn't see any misses. OK, I understand that much. But I'm not sure that disproves the .26667% theory yet. With your number of attacks, you were likely to have observed 2 misses, and yet you saw 0 misses -- that is definitely not helpful to the .26667% theory. But, on the other hand, I don't think it is fatal yet either, since 0 misses versus 2 misses is not a huge variance, given the relatively small sample size.

And then I got confused by the conclusion you reach when you switched to 350 hit rating. At 350 hit rating and 352 weapon skill, your chance to miss would be .2715% (i.e., under the .26667% theory). So, I think you should see a miss roughly every 368 attacks. Yet you suggest you should only see a miss every 20500 attacks, which is where I think I got lost a bit...

So, ultimately you reach the conclusion that weapon skill must equal at least .3707% -- which is the part I'm struggling with.

Sorry, if I've misread something. But I'm confused.

The 1 miss in 20500 attacks is under a 0.4% theory.
Under this, with 350 hit rating I should see 1 miss in exactly 20500 attacks, with 351 hit rating I should see none.

And as I've said, the data sample is quite low so far, but it is just so tedious to have to rely on other (stupid) players to be able to do 100 swings within 20-30 minutes. And if they are really stupid, it comes down to only 50 swings per 30 minutes.
I'm still continuing the test whenever I find time outside of raids and after work, but this is why I asked Karmon if he could do another of his raid tests with 344 hit rating and 3 weapon skill, since obviously he hasn't the problem I do, running around with 7000 hp and performing hemo with daggers and thus putting out crappy DPS, just to be able to test with such a high hit rating.

Karmon · 08/29/07, 5:14 AM

Originally Posted by sp00n

[..]
but this is why I asked Karmon if he could do another of his raid tests with 344 hit rating and 3 weapon skill,

I will try this friday on our Karazhan run with 343 and 344 hitrating.

Originally Posted by sp00n

since obviously he hasn't the problem I do, running around with 7000 hp and performing hemo with daggers and thus putting out crappy DPS, just to be able to test with such a high hit rating.

"Not a problem" is probably exaggerated

It hurts a lot. I loose about 230AP, the 2pc Netherblade bonus and the procs of DST and Abacus.