Forum Settings
       
Reply To Thread

Epona's on WSsFollow

#27 May 02 2011 at 12:58 PM Rating: Decent
**
983 posts
Other multi-attack things this may apply to that there should already be substantial data on:
OAT, Sea weapons, Saber Dance

* OAT was thought to be competitive.
* Fort Axe was thought to have DA/TA able to proc off Virtue stones procs, but with >3 rounds being demoted to 3.
* Saber Dance was thought to replace the /WAR Double Attack Job Trait (basically by me). (http://www.bluegartr.com/threads/100651-The-Current-Hierarchy-of-DDs?p=4262169&viewfull=1#post4262169 and further down the page)

Three questions, I guess:
1) What STR do I need to aim for on Hpmedes to hit for 0?
2) What is the minimum delay that KParser can reliably recognize DAs for? Ballpark is fine.
3) How would I write a script that uses Saber Dance every 3 minutes overnight and doesn't crash me?

3% back, 2% waist, 2% grip, 10% trait, 5% brutal, and 2% Aurore if I can = 24% DA from gear
20% average predicted DA from Saber Dance.

44% if they just add, 34% predicted from my previous tests.
____________________________
Yay for Jhereg!
http://www.ffxiah.com/player/Lakshmi/Byrth
#28 May 02 2011 at 1:30 PM Rating: Excellent
***
2,236 posts
Byrthnoth wrote:
Other multi-attack things this may apply to that there should already be substantial data on:
OAT, Sea weapons, Saber Dance

* OAT was thought to be competitive.
* Fort Axe was thought to have DA/TA able to proc off Virtue stones procs, but with >3 rounds being demoted to 3.
* Saber Dance was thought to replace the /WAR Double Attack Job Trait (basically by me). (http://www.bluegartr.com/threads/100651-The-Current-Hierarchy-of-DDs?p=4262169&viewfull=1#post4262169 and further down the page)

Three questions, I guess:
1) What STR do I need to aim for on Hpmedes to hit for 0?
2) What is the minimum delay that KParser can reliably recognize DAs for? Ballpark is fine.
3) How would I write a script that uses Saber Dance every 3 minutes overnight and doesn't crash me?

3% back, 2% waist, 2% grip, 10% trait, 5% brutal, and 2% Aurore if I can = 24% DA from gear
20% average predicted DA from Saber Dance.

44% if they just add, 34% predicted from my previous tests.


1) From what I remember, 61-62 str for the ones near the entrance; might be able to go up to 65 str for the ones at the back of the southern area, and possibly a few points higher with the ones in the northern area.

Edit: Found it -- post here, with max str for the various level hpemdes. 66 is max for the highest level mobs.

2) I'd guess 200-210 minimum, but would suggest 240 or higher if possible.

3) No idea.

34%: interesting prediction. 24% * (1-20%) + 20% * (1-24%) = 34.4%, using the xor calculations. Given this new understanding, that would imply that Saber Dance is explicitly different from DA, and suffers from the same problem where (effectively) if both proc at the same time, you get the benefit of neither.

Edited, May 2nd 2011 2:34pm by Kinematics
#29 May 02 2011 at 1:33 PM Rating: Decent
**
983 posts
Hmm... that may be a bad point to test then. In the past it has come out as a simple DA-10% from what is expected. At the time, I interpreted it as Saber Dance canceling out the DA trait from /WAR. My previous data doesn't exclude this model though.

If I get it working though, it won't matter. I can spend a few nights in sea and test a bazillion values.

Edited, May 2nd 2011 3:33pm by Byrthnoth
____________________________
Yay for Jhereg!
http://www.ffxiah.com/player/Lakshmi/Byrth
#30 May 02 2011 at 1:56 PM Rating: Excellent
***
2,236 posts
If it progresses in the expected 25%/20%/15% over 3 minutes..


10% DA from /war == 30% / 26% / 22%
Overall average: 26%

15% DA from /war + Brutal == 32.5% / 29% / 25.5%
Overall average: 29%

20% DA from /war + Brutal + Atheling + Twilight == 35% / 32% / 29%
Overall average: 32%

Gaining 3% avg DA for every 5% added in gear.

Avg if DA+SD-10%:
10%DA: 20% (if xor, expect parse to be 6% higher)
15%DA: 25% (if xor, expect parse to be 4% higher)
20%DA: 30% (if xor, expect parse to be 2% higher)



You'll need at least -15 str if you're hume dnc/war.

Acerbic Sash: -7
Hoard Ring: -4
Nemus Bazubands: -4
^- for base (need to take off Twilight; use Pole Grip for +2 DA instead)
Nemus Nails: -3
Snow Ring: -2
^- to compensate for Aurore body str, if used
#31 May 02 2011 at 4:11 PM Rating: Good
**
983 posts
Well, I'm a taru fortunately (75 STR base on 90DNC/45WAR without STR merits).

This post says I'll need 60 STR to bring fSTR to -1 on the highest level Hpmedes (67 AGI, 75 VIT, 275 or 280 Eva):
http://www.bluegartr.com/threads/68786-Dexterity-s-impact-on-critical-hits?p=3193007&viewfull=1#post3193007

I can probably avoid getting Nemus hands/feet synergized if I drop my STR merits and get Coeurl Ledelsens made (much easier).

Setup:
Cobra Staff/Pole Grip/____/Charis Feather
AF3+2 head/Love Torque/____/Brutal
AF3+2 Body/AF3+2/Hoard/Malflame
Atheling/Acerbic/Matre/Coeurl

DEX81+30 (dDEX=44 = 18% crit rate)
20% Double Attack
STR75-15 (fSTR=-1)
132 staff skill, +19 Acc from gear, +35 from traits, +55 from DEX = 241 Acc = ~55-58% hit rate

Tonight I will spam Box Step and staff with this setup, assuming I can find an Acerbic Sash and Coeurl Ledelsens. From there I can move on to testing Feather Step's potency, Step Acc gear, and other DA values.
____________________________
Yay for Jhereg!
http://www.ffxiah.com/player/Lakshmi/Byrth
#32 May 02 2011 at 5:08 PM Rating: Excellent
***
2,236 posts
A more complete str list is linked to in my post above (edited in, you may have missed it). The highest level om'hpemdes allow up to 66 str for your testing needs.

Since you're taru, then 80 str with merits would need just -14 str, which is doable with Acerbic+Hoard+Malflame; so, you shouldn't need to drop any merits, though by dropping 3 merits you add 2 more levels of mob you can work on.
#33 May 02 2011 at 5:41 PM Rating: Decent
**
983 posts
Masamune and pchan in the thread I linked seem to have shown experimentally that 60 or 61 is the upper limit on STR you can have and avoid hitting for >0. Specifically this post:
http://www.bluegartr.com/threads/68786-Dexterity-s-impact-on-critical-hits?p=3193007&viewfull=1#post3193007

Edited, May 2nd 2011 7:49pm by Byrthnoth
____________________________
Yay for Jhereg!
http://www.ffxiah.com/player/Lakshmi/Byrth
#34 May 02 2011 at 6:10 PM Rating: Excellent
***
2,236 posts
That's only for the ul's, which only go up to lvl 72. The om's go up to 77. See this post.
#35 May 02 2011 at 8:06 PM Rating: Good
Scholar
36 posts
So basically (based on pchan's results) from an average hits/round point of view, if your TA rate > 33%, any DA will mess up your output, or if your DA rate > 66%, any TA will mess up your output. But if your TA < 34% and DA < 67%, that is ok (unless I horribly screwed this up).

Are DA > 66% or TA > 33% even possible? I don't recall.
#36 May 02 2011 at 8:59 PM Rating: Excellent
***
2,236 posts
Antisense wrote:
So basically (based on pchan's results) from an average hits/round point of view, if your TA rate > 33%, any DA will mess up your output, or if your DA rate > 66%, any TA will mess up your output. But if your TA < 34% and DA < 67%, that is ok (unless I horribly screwed this up).

Are DA > 66% or TA > 33% even possible? I don't recall.



DA > 66% isn't possible. TA > 33% is.

All values of either will mess with the value of the other, with higher values causing progressively greater devaluation of the other stat. I've already worked out that TA > 20% means that Raider's Earring (1% TA) beats Brutal Earring (5% DA) for any reasonable amounts of DA (anything up to ~20%).

There is no solid rule of thumb yet for the preference in relative values. We still need to compare a lot of different scenarios to see the relative impacts of each stat in gear choices.
#37 May 02 2011 at 11:06 PM Rating: Good
**
712 posts
Kinematics wrote:
Antisense wrote:
So basically (based on pchan's results) from an average hits/round point of view, if your TA rate > 33%, any DA will mess up your output, or if your DA rate > 66%, any TA will mess up your output. But if your TA < 34% and DA < 67%, that is ok (unless I horribly screwed this up).

Are DA > 66% or TA > 33% even possible? I don't recall.



DA > 66% isn't possible. TA > 33% is.

Wouldnt that only apply to thf, since as far as I know, theyre the only job that can reach 30%+ TA with gear/traits/merits/atmas. Other jobs would be stuck around the 15% ~ 28% no?
____________________________
SERVER: IFRIT
WHM/BRD: Triangulum
PLD
MNK
SAM
BLM
DRK

#38 May 03 2011 at 12:40 AM Rating: Good
Sage
**
523 posts
hitoseijuro wrote:

Wouldnt that only apply to thf, since as far as I know, theyre the only job that can reach 30%+ TA with gear/traits/merits/atmas. Other jobs would be stuck around the 15% ~ 28% no?


Any job could sit at a 35% Triple Attack rate with proper atma support: Atma of the Apocalypse for 15%, Atma of the Lion and Atma of Alpha and Omega for 10% each. Don't why you'd ever want to do that, but it is possible if not feasible.
____________________________
Server - Bahamut
Name - dawdr
Mithra 75thf/40rng/37nin/37war/37drk/37sam
#39 May 03 2011 at 3:55 AM Rating: Good
Scholar
36 posts
Kinematics wrote:
All values of either will mess with the value of the other, with higher values causing progressively greater devaluation of the other stat. I've already worked out that TA > 20% means that Raider's Earring (1% TA) beats Brutal Earring (5% DA) for any reasonable amounts of DA (anything up to ~20%).

There is no solid rule of thumb yet for the preference in relative values. We still need to compare a lot of different scenarios to see the relative impacts of each stat in gear choices.


Well, yeah. All I mean is that combining DA and TA will not be detrimental (up to a point), like DA doesn't "nerf" Ridill. Tradeoffs are a different issue (as you recall circa 2007 you could still use Brutal with Ridill since it's not like there was really anything better to use). For THF, /WAR doesn't seem to be very good vs. /NIN if you're sitting on 30% TA for example.

Edited, May 3rd 2011 5:56am by Antisense
#40 May 03 2011 at 9:16 AM Rating: Decent
**
415 posts
62 str works on the hpende that roam in the crag entrance areas.
#41 May 03 2011 at 9:39 AM Rating: Excellent
***
2,236 posts
Antisense wrote:
Kinematics wrote:
All values of either will mess with the value of the other, with higher values causing progressively greater devaluation of the other stat. I've already worked out that TA > 20% means that Raider's Earring (1% TA) beats Brutal Earring (5% DA) for any reasonable amounts of DA (anything up to ~20%).

There is no solid rule of thumb yet for the preference in relative values. We still need to compare a lot of different scenarios to see the relative impacts of each stat in gear choices.


Well, yeah. All I mean is that combining DA and TA will not be detrimental (up to a point), like DA doesn't "nerf" Ridill. Tradeoffs are a different issue (as you recall circa 2007 you could still use Brutal with Ridill since it's not like there was really anything better to use). For THF, /WAR doesn't seem to be very good vs. /NIN if you're sitting on 30% TA for example.


For TA > 33%, adding Brutal will start to reduce your average hits per round, so that's the point where you're actually hurting yourself adding DA. At TA = 30%, adding Brutal is a 0.3% gain in hits per round (assuming no other DA at all). At TA = 20%, adding Brutal is a 1.4% gain in hits per round. At TA = 10%, adding Brutal is a 2.9% increase in hits per round. At TA = 0%, adding Brutal is a 5% gain in hits per round.

So the loss, at 10% increments of TA, as a percentage of the previous gain amount is: 42%, 52%, 79%. The loss accelerates as TA increases, but is fairly substantial even at low levels. In the 10%-20% TA range (Apoc atma and perhaps Epona's Ring), you can expect all DA gear to have about half its listed value in actual improvement to hits per round.
#42 May 03 2011 at 11:21 AM Rating: Good
**
415 posts
A big part of this is due to the fact that the base attack per round is a lot bigger than 1, though, and not ompletely to the new findings.

For instance with apoc, epona, atheling as base, sub ninja :

previous model :
base attacks/round : 1+1*0.06+2*0.18*(1-0.06)=1.3984
with brutal earring : 1.4304

=> net gain = +2.28%

Current model :
base : 1+1*0.06*(1-0.18)+2*0.18*(1-0.06)=1.3876
with brutal : 1.4106

=> net gain : +1.66%

Also we have to keep in mind that the global dps is not a function of the average round/sec




Edited, May 3rd 2011 1:21pm by lynnminmay
#43 May 03 2011 at 4:32 PM Rating: Decent
**
983 posts
I can't get anyone to sell me a Acerbic Sash, so I ate some cursed soup and did a test run to make sure my script would work and to get some preliminary data (it didn't, but I think I can fix it). 65 STR does work on the highest level ones at the crags.

What I found, on the other hand, was thoroughly weird.

24% double attack rate (461/1935) @3 hour mark with Saber Dance up every 3 minutes and 22% DA through traits and gear. The script failed once and it parsed for ~1 minute without refreshing Saber Dance, but then I did it manually and got it back on track. The only expaination I have is that Saber Dance is somehow affected by accuracy.

I need a damned Acerbic Sash so I can test this overnight.
____________________________
Yay for Jhereg!
http://www.ffxiah.com/player/Lakshmi/Byrth
#44 May 03 2011 at 4:52 PM Rating: Excellent
***
2,236 posts
Not sure what happened there, but I'm thinking you need a baseline value first. That is, dnc/nin with no DA gear at all, running the SD script overnight, so you know what to expect from the script's performance. Then do another run to see the impact of adding the DA gear.
#45 May 03 2011 at 5:06 PM Rating: Decent
**
983 posts
I agree, but before I do that I will need to get an Acerbic Sash.
____________________________
Yay for Jhereg!
http://www.ffxiah.com/player/Lakshmi/Byrth
#46 May 04 2011 at 3:27 AM Rating: Good
**
415 posts
Possible explanation : it doesn't overwrite itself (lol useless merits ?), it works like umpetus and you need to hit the target, if so, I'll suggest using D1 daggers or lady bell. It could be that it is just an oat that only procs if the hit lands and doesn't double attck (this would be easy to check with no DA and a few hits).
#47 May 04 2011 at 5:24 AM Rating: Decent
**
983 posts
lynnminmay wrote:
Possible explanation : it doesn't overwrite itself (lol useless merits ?), it works like umpetus and you need to hit the target, if so, I'll suggest using D1 daggers or lady bell. It could be that it is just an oat that only procs if the hit lands and doesn't double attack (this would be easy to check with no DA and a few hits).


Unfortunately, it's not an accuracy check on the first swing. After 6 minutes I had 5 possible "Zanshins" with less than a 100% hit rate on them. If it's somehow accuracy related, it's not in an obvious way.

I agree that it might not overwrite. Well, I was thinking that potentially re-using Saber Dance might just extend duration unless you cancel it. I have the script fixed now and will let it run while I'm at work. If this comes out with an unexpectedly low %DA, I'll add "Cancel" before Saber Dance activation and see if that changes anything.

Edit:
CDF, can you think of a reason kparser might be missing attacks? Specifically after doing "Stop" and then "Continue"

Edited, May 4th 2011 8:31am by Byrthnoth
____________________________
Yay for Jhereg!
http://www.ffxiah.com/player/Lakshmi/Byrth
#48 May 04 2011 at 10:10 AM Rating: Excellent
***
2,236 posts
Actually, that's an interesting idea, and you can check if, in fact, your DA rate is decaying continually, and reuse of Saber Dance doesn't reset the DA values, with your existing parse.

What you can do is run a Split on the parse in 1 minute chunks, and look at the DA rate within that period. While there's liable to be chopped results due to attack rounds crossing over 1 minute boundaries, you should at least be able to see if there's any general trends.

If you want to get really fancy, you can use the details section of the Extra Attacks tab to pick out more precise time values to split on so that you don't have attacks overlapping selected boundaries. Realize that to include all attacks for one multiattack round you should set the split to end immediately before the next round begins, as the attacks that are grouped into each round will be spread out over a period of time.

If the DA rate progresses in sets of high/medium/low/high/medium/low, then SD properly overwrites and resets itself. If it progresses as high/medium/low/low/low/low, then the decay is based on the initial start time of the JA, and may explain the long-term results you're seeing (24% overall with 22% in gear/traits, where the vast majority of the parse got no DA boost).



Edited, May 4th 2011 11:13am by Kinematics
#49 May 04 2011 at 10:24 AM Rating: Decent
*
134 posts
Quote:
and as a matter of fact, I don't even recall saying anything to be a **** here. I asked you why you would use a twilight belt, since you were counting it for DA.

That was a serious question (one that you haven't answered yet).


A mnk with no BB could use twilight and keep haste cap and get the extra DA value from the belt, sure it would work out to better DoT than the +8 str or whatever the second mod on brown is.

Just a thought.
#50 May 04 2011 at 10:56 AM Rating: Excellent
***
2,236 posts
Revision to above suggestion: Just use the Extra Attacks details listing and do a manual count per minute range. It'd be faster and easier.
#51 May 04 2011 at 2:26 PM Rating: Decent
**
983 posts
How do I quote the parse data like pchan did?

All intervals are just 95% confidence

4348 rounds, 811 DAs, 18.65% average DA rate (+/- 1.16%)

DA rate was unaffected by missing the first swing, and the hit rate of DAs was not significantly different than normal hit rate (58.81% +/-1.34%)

Box Step hit rate was 62.43% +/- 2.42%, so steps either have a modest accuracy bonus over normal hits or double attack can proc on them. 62.43/(1+.1865*.5881) = 56.26, and the way the script is set up there are actually more box steps later in the DA cycle, iirc.

Crit rate was 14% +/- 1%

It's going to take me a bit to count the 3 minute blocks by hand. First I'm going to **** around trying to get it into matlab with space delimiters to see if I can make a pretty graph of it.

Okay, I can move all the extra attacks into matlab and turn the timestamp from the extra details into a timestamp (number in seconds) and pull out the number of attacks at that timestamp. Then I subtract off the offset (started at 8:30ish) and divide by my cycle time (181 seconds) to make it cyclical. From there I subtract off the whole number (so I'm just left with a fraction out of 181) and sort high to low. Now I'm attempting to sum the number of hits in every "second window," and I'm semi-convinced my code is working now <_<.

http://img854.imageshack.us/img854/4285/unsmoothed.jpg

http://img195.imageshack.us/img195/7229/boxavg.jpg

The first is unsmoothed and relative to position in the cycle. It's rough because there are only about 25 samples for each point. Despite the obvious spike at the beginning (and end), I'm not entirely sure where along the cycle I used Saber Dance. Obviously I'd guess the beginning, but I don't really know.

The second is smoothed (average over 10 bins, relative to the frequency in each bin), but there's still no guarantee that every point is equally represented. I think maybe I just need more sample size, so I'm letting it run another 10 hours.

Edited, May 4th 2011 6:01pm by Byrthnoth
____________________________
Yay for Jhereg!
http://www.ffxiah.com/player/Lakshmi/Byrth
Reply To Thread

Colors Smileys Quote OriginalQuote Checked Help

 

Recent Visitors: 1 All times are in CDT
Anonymous Guests (1)