[Rogue/Warrior] Weapon Skill Adjustment Discussion - Page 21

Karmon · 08/03/07, 4:51 AM

Originally Posted by Olgas

OK, ran Gruul's tonight with 8.624% +hit (136 hit) and no +Weapon Skill.

As expected, this test confirmed that the base miss rate for Ranged DPS against a Level 73 is greater than 8.6%.

Here were my results for the evening:

Auto Shot: 441 hits, 127 crits, 3 misses (.555% miss)
Steady Shot: 422 hits, 130 crits, 4 misses (.719% miss)
Arcane Shot: 8 hits, 0 crits, 0 misses (0% miss)

Total Shots: 1105 Total Misses: 7 (.633% miss)

So far, Ranged DPS is exhibiting the exact same characteristics as melee DPS. I don't know how to do that 95% confidence range calculation, but it seems to me like this test again supports a base miss rate of around 9% (9.23%)

Even though this sample size was not that big (only 1105 shots), I don't feel the need to repeat it. The important thing is that there were misses at all.

Doing the numbers we get an interval (adjustet with the 136 hitrating base): [8.789% - 9.725%]
The total swings is a bit low thats why the interval is so wide.

Originally Posted by Olgas

Meanwhile, I thought I'd also share the results of my pet attacks from tonight's Gruul's run:

Melee: 1339 attacks, 58 misses (4.332%)
Specials (all combined): 919 attacks, 42 misses (4.570%)

Total for Pet: 2258 attacks, 100 misses (4.429%)

As a reminder, my pet has +4% hit from talents. So this test suggests a 8.429% miss rate for the pet, which is a bit lower than our expected 9%.

Well either pets follow a different mechanic (or same mechanic with different constants) or our assumption about petattacks are wrong or your pet has a hidden +hit or +skill bonus

Fugazor · 08/03/07, 12:38 PM

Originally Posted by Dontmindme

Are you sure the armory gear is what they used in the raid? Did you have a Dranei aura? Did you have Improved Faerie Fire up? Did they buff with a Fire-Toasted Bun?

1) yes
2) horde so no
3) no boomkins so no
4) 99.9% no

I thought 8.6/27.6% was correct until this post:
http://elitistjerks.com/424269-post120.html

Basically people say that 360 ws + 308 hr + 5/5 precision = never miss for rogue:
19.5% (308 HR) + 5% + 0.4% (10 weapon skill) ~ 25%

What if hunters in TBC have diffrent miss rate from 2h warriors/pala and/or feral druids? I checked quickly this thread but could not find any WWS raid data confirming any theory.

As for dual-wield, I had 360 ws and ~22.8% (17.8+5) hit and results from WWS are:
VR: 626 attacks, 6 miss (0.96%)
Lurker: 781 attacks, 6 miss (0.77%)
Morogrim: 1096 attacks, 16 miss (1.46%)
Tried to pick fights with none to almost none adds. It shows that mi miss is around 24-25%. Now ok, there can be some error, some luck but with 28% miss it should be 3-4% hit from luck on those fights. Sorry but I could not believe in THAT much of luck.

Of course there can always be possibility that WWS is bugged somehow.

Karmon · 08/03/07, 1:12 PM

Originally Posted by Fugazor

I thought 8.6/27.6% was correct until this post:
http://elitistjerks.com/424269-post120.html

Basically people say that 360 ws + 308 hr + 5/5 precision = never miss for rogue:
19.5% (308 HR) + 5% + 0.4% (10 weapon skill) ~ 25%

That is correct as far as we know.
The small little detail that always gets overlooked is the 360 skill.

A few post above Spoon derived two formulas based on the testing here.
Basemiss:
5+dualwield + (bossdefence - (350+weaponskill))*0.1 if (bossdefence - your weaponskill) <= 10)
5+dualwield + 10*0.1 + (bossdefence - (350+weaponskill) - 10)*0.6 if (bossdefence - your weaponskill) >= 10)

So for someone with only 350 weaponskill against a boss that translates to 28% for dualwielders and 9% for all others.

The numbers are not rock solid yet but pretty convincing

Olgas · 08/03/07, 1:22 PM

Originally Posted by Fugazor

What if hunters in TBC have diffrent miss rate from 2h warriors/pala and/or feral druids? I checked quickly this thread but could not find any WWS raid data confirming any theory.

As for dual-wield, I had 360 ws and ~22.8% (17.8+5) hit and results from WWS are:
VR: 626 attacks, 6 miss (0.96%)
Lurker: 781 attacks, 6 miss (0.77%)
Morogrim: 1096 attacks, 16 miss (1.46%)
Tried to pick fights with none to almost none adds. It shows that mi miss is around 24-25%. Now ok, there can be some error, some luck but with 28% miss it should be 3-4% hit from luck on those fights. Sorry but I could not believe in THAT much of luck.

If you have 360 Weapon Skill equipped, then it is as though you are attacking a Level 71 mob when you are attacking Level 73 mobs. Which means you are facing an opponent with a base miss rate of 24.5%. So your results are perfectly in line with our working theory.

The ~28% base miss rate applies if your Weapon Skill is at 350, attacking a mob with a 365 Defense Skill.

Also, it is still theoretically possible that Hunters have a different combat calculation from other classes. But it seem pretty unlikely. So far, all the testing has indicated that Ranged physical DPS works exactly the same as Melee physical DPS (at least with respect to calculating hits and misses), and so that is our working assumption right now.

Also, you have to be very careful to not pollute data tests by including combat with adds. Bosses are level 73 and so have a base miss rate of ~28%. Adds on the other hand, will usually be Level 71 or 72, so they will only have a miss rate of 24.5% or 25%, respectively. If your test mixes boss fight with add fights, then you will compromise the integrity of the test.

Olgas · 08/03/07, 1:39 PM

Originally Posted by Karmon

5+dualwield + 10*0.1 + (bossdefence - (350+weaponskill) - 10)*0.6 if (bossdefence - your weaponskill) >= 10)

...

The numbers are not rock solid yet but pretty convincing

Just to clarify, we still haven't really tested whether those first five point of Weapon Skill (i.e., from 350 to 355) reduce miss rate linearly (i.e., by a an equal .6% per Skill level).

I certainly think it is the more elegant formula. (And I actually hope it is right!)

But there is still this competing idea:

Base miss rate is:

24% + (Defense Skill - Weapon Skill)*0.10 [if (Def Skill - Wep Skill) <=10]
24% + (Defense Skill - Weapon Skill)*0.26667 [if (Def Skill - Wep Skill) > 10]

Which implies an abrupt and fairly large reduction in miss rate when going from Weapon Skill 354 to 355.

As I said before, I certainly emotionally favor Sp00n's formula over the Wowwiki formulation. But I don't want to assume it is accurate before we have tested the impact of incremental Weapon Skill between 350 and 355. Right now we don't actually have any data to favor one theory over the other.

Arcazua · 08/06/07, 1:13 AM

Did a test of my own today. I will hopefully follow it up with a more pertinent one.

Blizzard has stated that bosses behave like +3, and we are all assuming this is still true. However, that information came out before TBC. Since there are no standard level 73 mobs to test, this means using a lower-level alt to check.

http://img.photobucket.com/albums/v3...MissProof1.jpg
http://img.photobucket.com/albums/v3...MissProof2.jpg

Those are two pictures of a test I ran with a level 52 hunter against a level 56 BL servant. Note that this is a +4 mob, and yes I'm aware we really wanted a +3, but I couldn't find an L55 and didn't want to level just so I could work on my weapon skill some more.

The hunter has 260 bow skill (non-troll), which is max for 52, and no hit rating at all. The pet is 51, and you'll notice some of the pet's specifics mentioned in the one screenshot as well -- "miss" also includes parry and dodge and stuff, though, so I'm not sure if anyone really cares.

However for the hunter safely shooting for afar, with 2000 autoshots taken, there were a total of 178 misses, or 8.9%. I would hope we can consider 2000 statistically significant, but stat was never one of my finer points and we are effectively looking for accuracy up to 3 digits, so maybe not. It did start closer to 8.7% and crept up to 9.0 in the last 500, and then fluctuated a little. 178/2000 is pretty darn close to 8.9%, though, so I'm going to label this the miss rate for a standard player against a +4.

The special attacks I wove in as mana permitted had a higher miss rate. They are not statistically significant, although if you incorporate them into the autoshot, we then have a 9.1% miss rate. Do with that what you will, but I am going to assume that we do not know for certain that specials operate the same as base attacks, so I'm going to stick to the 8.9 figure.

Now, with our current model(s) that we've been using, 8.9% is less than what we would expect for a +4, (as is 9.1,) and even less than Olgas is theorizing is the miss rate for bosses, which means one or more of a few things:
1) Lower-level targets have a different formula used than level 70. (Unlikely considering they'd use the pre-TBC model then, which also should be higher -- about 9.8%)
2) The boss miss rate is actually under 9%.
3) Bosses are not true +3's for the sake of physical attacks.

If you trust Olgas's tests, then I suspect #2 is false. Although (s)he hasn't done testing on exactly 9% +hit, Olgas has at least found that boss miss rate is higher than the old original 8.6%, and 9% is what I suspect will be the agreed-upon figure. If that much is true, and my test is reasonably accurate, then #3 very well might be correct.

If and when I do the test with a +3, I'll be sure to report back. 75+ minutes of autoshooting and hitting Mend Pet every 15 seconds is zzzz, though.

Gruntle · 08/06/07, 4:19 AM

Originally Posted by Arcazua

However for the hunter safely shooting for afar, with 2000 autoshots taken, there were a total of 178 misses, or 8.9%. I would hope we can consider 2000 statistically significant, but stat was never one of my finer points and we are effectively looking for accuracy up to 3 digits, so maybe not. It did start closer to 8.7% and crept up to 9.0 in the last 500, and then fluctuated a little. 178/2000 is pretty darn close to 8.9%, though, so I'm going to label this the miss rate for a standard player against a +4.

Nice testing, however, 2000 shots with 178 misses means that the 95% confidence limits are given by [7.65, 10.15]. You need very large samples to get confidence limits down to the third digit.

Your test does suggest that the Spoon/Olgas/Karmon theory above fails for +4 mobs. The miss rate vs +4 mobs according to that theory should be 12% (5 + 10*0.1 + 10*0.6). You didn't have Surefooted talent with the hunter? (should give +3% to hit I think?). That would make your tests fall exactly in line with the theory.

If you don't have the talent, I guess there's another complicated effect kicking in at +4 level difference. Or, bosses might behave differently.

Karmon · 08/06/07, 5:01 AM

Originally Posted by Gruntle

Nice testing, however, 2000 shots with 178 misses means that the 95% confidence limits are given by [7.65, 10.15]. You need very large samples to get confidence limits down to the third digit.

Your test does suggest that the Spoon/Olgas/Karmon theory above fails for +4 mobs. The miss rate vs +4 mobs according to that theory should be 12% (5 + 10*0.1 + 10*0.6). You didn't have Surefooted talent with the hunter? (should give +3% to hit I think?). That would make your tests fall exactly in line with the theory.

If you don't have the talent, I guess there's another complicated effect kicking in at +4 level difference. Or, bosses might behave differently.

A possible alternative is, that they clamp the leveldifference at +3 level. So any mob >= 3 level counts as 365 defence.

songster · 08/06/07, 6:27 AM

Originally Posted by Karmon

A possible alternative is, that they clamp the leveldifference at +3 level. So any mob >= 3 level counts as 365 defence.

Easy to test - get a warlock to summon a level 1 character to SMV and have a go at one of the channelers.

Karmon · 08/06/07, 6:36 AM

Originally Posted by songster

Easy to test - get a warlock to summon a level 1 character to SMV and have a go at one of the channelers.

I would suggest a character with higher level than 30 or so.
There are a number of exceptions to mechanics if the character level is below a certain threshold.

Arcazua · 08/06/07, 1:57 PM

Originally Posted by Gruntle

Nice testing, however, 2000 shots with 178 misses means that the 95% confidence limits are given by [7.65, 10.15]. You need very large samples to get confidence limits down to the third digit.

Sigh, yes. Someone is going to have to remind me how to find my own confidence intervals. I'd settle for 90% and/or accurate to within a quarter percent rather than a tenth.

You didn't have Surefooted talent with the hunter? (should give +3% to hit I think?). That would make your tests fall exactly in line with the theory.

No, went pure BM to get to Serpent's Swiftness so the test would go by faster. I didn't want to pick up any talents or effects that could throw off the testing. There was absolutely no chance-to-hit modification on this character.

Originally Posted by songster

Easy to test - get a warlock to summon a level 1 character to SMV and have a go at one of the channelers.

Even easier: get a level 1 hunter into Duskwood and shoot at Stitches while he fights the watchers. (Has to be a hunter or someone with a ranged attack since he has a short-range AE aura.)

Karmon, while it's true that a level 1 can hit a level 5 mob alright, I believe the primary exceptions for low-levels are with regard to PvP.

Olgas · 08/06/07, 2:18 PM

Originally Posted by Gruntle

Nice testing, however, 2000 shots with 178 misses means that the 95% confidence limits are given by [7.65, 10.15]. You need very large samples to get confidence limits down to the third digit.

Your test does suggest that the Spoon/Olgas/Karmon theory above fails for +4 mobs. The miss rate vs +4 mobs according to that theory should be 12% (5 + 10*0.1 + 10*0.6).

Remember, that we still have two competing theories for how the formula works when Defense Skill exceeds Weapon Skill by more than 10:

The "Sp00n/Karmon" formulation:
5 + 10*0.1 + (Def Skill - Weap Skill - 10)*0.6

And the "modified Wowwiki" formula:
5 + (Def Skill - Weap Skill)*.266667

We all like the Sp00n/Karmon formulation better because it is cleaner and yields a linear impact between Weapon Skill 350 and 355. But we have no actual data yet to prefer one formula over the other yet.

And, actually, if you apply the modified Wowwiki formula to this particular test against a +4 mob, you get an expected miss rate of 10.33% (instead of 12%) which is still outside the 95% confidence level of the test... but just barely now, since the high end is 10.15%.

Gruntle · 08/07/07, 4:57 AM

That is of course true Olgas. Didn't think of that.

If i understood the modified WoWwiki theory correctly you would use the 0.2667 factor when the difference is 10 or greater. The problem is ofc (as you say) that the function is not continous at the difference 10 boundary, it has a jump at that point. It may be that's how blizzard constructed it, but to a scientific mind it just feels very wrong

Will start raiding with my dps warrior really soon, so will try to parse some logs myself. More data sets can never be wrong (carefully checked for systematics ofc...)

Olgas · 08/10/07, 10:23 AM

Originally Posted by Olgas

OK, ran Gruul's tonight with 8.624% +hit (136 hit) and no +Weapon Skill.

As expected, this test confirmed that the base miss rate for Ranged DPS against a Level 73 is greater than 8.6%.

Here were my results for the evening:

Auto Shot: 441 hits, 127 crits, 3 misses (.555% miss)
Steady Shot: 422 hits, 130 crits, 4 misses (.719% miss)
Arcane Shot: 8 hits, 0 crits, 0 misses (0% miss)

Total Shots: 1105 Total Misses: 7 (.633% miss)

So far, Ranged DPS is exhibiting the exact same characteristics as melee DPS. I don't know how to do that 95% confidence range calculation, but it seems to me like this test again supports a base miss rate of around 9% (9.23%)

Even though this sample size was not that big (only 1105 shots), I don't feel the need to repeat it. The important thing is that there were misses at all.

For my next test, I'm going to jump up to exactly 9% (142 hit) and see if I see misses at that point. Unfortunately, I won't get the chance to do this in a raid until Monday...

Unfortunately, I've only been able to go on about a raid and a half over the last week. So, I've had very little opportunity to do this testing.

However, last night we did do a quick Gruul's run. And I was equipped with exactly 142 hit gear (9.00%).

728 hits, 168 crits, 0 misses

So, zero misses in 896 shots.

This is too small a sample size to draw any definitive conclusions from. I figure I probably need to see something like 5000 shots without a miss, before we can be comfortable about the 9% cap. But I wanted to share these preliminary results with you.

Ashlan · 08/10/07, 12:42 PM

slightly offtopic .. but in order to help us "not-so-statistics-pros" in testing ....

could someone give me the numbers, how many samples are necessary to get a 95% confidence interval for
+/- 0.5% ... 1% range
+/- 0.1% ... 0.2% range
+/- 0.05% .. 0.1% range

Ashlan