Faith vs Spharai Parse data? Anyone have?

[Kirschy]

It's a fairy easy test. I started running it a while ago when testing the delay from WS/JA/etc...

It's FRAPs test, so the accuracy is fairly precise.

I never got a chance to complete the testing, but I remember it being a base delay of like 60? I've been fairly busy lately, but if I can get a chance I'll repeat the tests and post the data for you all~

[Motenten]

Actually, that little discussion brings up an interesting side issue if you're going to try to get the weaponskill/JA delay stuff.

For many years it's been assumed that delay is 60 per second. I did some of the original tests to get that number, but that was basicly just a bunch of stopwatch-timed sets of attacks. The actual value I got was closer to 59 delay per second, but I figured it was just margin of error.

Well, guess what? The Zanshin tests I'm doing give me something far more accurate. And what I get is -not- 60 delay per second, but closer to 58.5.


Weapon: Ceremonial dagger -- delay 183
Gear: 5% Haste (Turban)

Expected final delay: 174 or 175, depending on if 5% haste is 50/1024 or 51/1024.


Current pld/sam test:
6666 attack rounds (no parser error rounds) in 19,843 seconds.
Delay @ 60 delay/second: 2.9767 * 60 = 178.6

First sam test:
25410 attack rounds (with 7 additional rounds from parser error) in 75,767 seconds.
Delay @ 60 delay/second: 2.9810 * 60 = 178.9

The sample size is large enough that those results are accurate to within about 1 delay (half that for the second test). Even with that, it's nowhere near the expected delay.

Using the numbers to back-figure the delay/second value, you end up with 58.45 and 58.37 delay per second (with a small margin of error), respectively, if the actual delay is 174, and 58.79 and 58.71 delay per second if the actual delay is 175.


Unfortunately this is not a truly clean test, since I have haste gear on. However given that haste has been pretty accurately nailed down in other testing, I don't think it likely that the hasted delay is inaccurate.

I also have no way to explain why such a peculiar number would be chosen for the delay:seconds ratio.


In any case, if these results are indeed valid then you'll need to account for them when trying to sample weaponskill delay.

For the weaponskill delay test, the way I would do it would be to get a set of 3 people: someone set up to use the d0 weapon (someone without any DA/TA/Zanshin, preferably), someone who can sleep the mob, and a sam with capped Shikikoyo. Perform normal attacks for a stretch, and have the sam shiki the attacker periodically. When the attacker has TP, weaponskill (from menu), followed by having the mage sleep the fish. The attacker can continue to attack without waking it up.

Proceed through a series of weaponskills that way, then get the total attack count and delay that the test covered and compare it with expected values.

[Kirschy]

Yea, FRAPS testing shows this discrepency too. I always measured in frames and converted to estimated delay. There's a good chance delay truncates at some point too. I always need to do a baseline to test so that i can accurately compare the two.

FRAPS testing will work for delay on JA/WS, because you record a baseline and then compare the baseline to the change. It's highly accurate, and is fairly fast to perform~

[Kirschy]

Was looking up some old tests, so I could find what the FRAPS test showed for frames->delay:

http://www.bluegartr.com/forum/58770-slow-hojo-slow-ii-2.html

Huge Hornet:
181.3
181.4
181.3
181.0
181.7
182.0
181.5
181.6
------
Avg~181.5

Carnage Elegy:
271.8
272.2
271.0
272.2
271.9
-----
Avg~271.82

271.82/181.5 = 1.498 (49.8% Slow)

In general, 1 frame = 1/30th second.
So assuming Bees have a delay of 350...
My test showed 181.5 frames / 30 = 6.05 seconds.
350 delay/6.05 seconds = 57.85 delay/sec.

It's not as precise as your test, but it's very close in accuracy and much quicker to perform~

[Shamaya]

Interesting..

Forgive me, Kirschy. This is probably very obvious to you. But I want to make sure I'm understanding correctly. So you use fraps and then load the video. The frame where you first see the hornet attack in the log, this is t1. The frame where you see the next successive attack in the log, this is t2. The total frames from t1 to the frame just before t2 is your delay in frames. From there you can convert into delay in units of real time and delay in units of game-measured delay. And of course you compared known game delays the total number of frames in a video, as well as the time length of the video to come up w/ the ratios. Is this how you did it?

I didn't follow the haste testing closely at all. So if that's how you did it, then for the weapon skill delay, we would do something like: Take t1 when you see a round of attacks in the log or tp gained on your bar, just before weapon skilling. Take take t2 when you see the next round, after weaponskilling. Convert frames to delay. Subtract your known melee delay/round from this number. This is the delay caused by weaponskilling. Likewise, the same could be done for JA pauses, etc, as in spell times and the values found in your haste studies.

Is this correct?

[pchan]

Totally different subject but does the crit +6% proc on Double attacks and / or counter ?

[CDF]

This might be interesting mainly for those who want to do precise calculations, but you can account indirectly for "lossy TP" by computing exactly the average number of rounds to reach 100 TP, given DA, whether dual wielding, etc. I made a spreadsheet a while ago as a proof of concept. (It doesn't account for monk kick attacks or virtue weapons yet.)

Most of it should be fairly simple to figure out. For example, suppose you're interested in the exact increase in average WS frequency from going from a 6 to 5 hit, given 15% DA. The paper-napkin observation is that it will be a 25% increase in frequency (after a WS), but actually it will be (5.605411489/4.690087009 - 1)*100% = 24.3% (in the long run).

You can also figure out the proportion of hits in excess of 100 TP to account (approximately) for the consequences of excess TP, such as loss of latent effects when > 100 TP and increase in average WS damage from TP in excess of 100 TP, if you really wanted to go that far. For example, given 15% DA, if it takes 6 hits to get to 100 TP (starting from 0 TP), on average the proportion of hits that will occur in excess of 100 TP, when spamming WS at 100 TP, is 6.123912051/6 - 1 = .02065.

[Kirschy]

Interesting..

Forgive me, Kirschy. This is probably very obvious to you. But I want to make sure I'm understanding correctly. So you use fraps and then load the video. The frame where you first see the hornet attack in the log, this is t1. The frame where you see the next successive attack in the log, this is t2. The total frames from t1 to the frame just before t2 is your delay in frames. From there you can convert into delay in units of real time and delay in units of game-measured delay. And of course you compared known game delays the total number of frames in a video, as well as the time length of the video to come up w/ the ratios. Is this how you did it?

I didn't follow the haste testing closely at all. So if that's how you did it, then for the weapon skill delay, we would do something like: Take t1 when you see a round of attacks in the log or tp gained on your bar, just before weapon skilling. Take take t2 when you see the next round, after weaponskilling. Convert frames to delay. Subtract your known melee delay/round from this number. This is the delay caused by weaponskilling. Likewise, the same could be done for JA pauses, etc, as in spell times and the values found in your haste studies.

Is this correct?

Fairly close to what I did. I used 5-20 (usualli ~10) melee attacks to get an average. Like you said, I loaded a fraps'd video and recorded the exact frame a melee animation started, and did this for every attack around. I should have posted one of my tests using a melee weapon of known delay, not a mob for which I need to guess delay. But I can do that later today~ (It shows a similar pattern.)

For testing JA just need to record ~10 sets of melee round but using voke inbetween each one, and calculate the difference. For WS, setting up a 6-hit build and recording for enough melee attacks for 5 weapon skills would be sufficient to show a small delay.

The testing was supposed to cover more than just JA/WS/Item/Magic delays, and covered the tests in low delay, high delay, low haste, high haste situations. I just never completed the testing, and got sidetracked to other projects. When I get a chance, I'll run a smell test and give you an estimate on it, so that the item comparison and be more accurate~

[Appie]

Isn't this a misnomer? Even with high Subtle blow gear on, the amount of TP fed to get to 100 TP is worse than melee two handers, no?

God, this reminds me of 2005 where I had to show my LS that MNK isn't the tp feeder they claim it to be and that their GAxe WAR/NINs (SAM wasn't the bandwagon back then) were feeding it more TP or the same as a MNK would (with a black belt and rajas as basic subtle equip). The subtle blow trait got not love back then, which THF should have gotten especially back then with their shitty pre-dagger buff for daggers over lv66, such tp feed for crap dmg XD

[Vodou]

God, this reminds me of 2005 where I had to show my LS that MNK isn't the tp feeder they claim it to be and that their GAxe WAR/NINs (SAM wasn't the bandwagon back then) were feeding it more TP or the same as a MNK would (with a black belt and rajas as basic subtle equip). The subtle blow trait got not love back then, which THF should have gotten especially back then with their shitty pre-dagger buff for daggers over lv66, such tp feed for crap dmg XD

I've only thought about this issue casually, and atm I am recovering from a minor surgery induced drowsiness, but I always thought these arguments resembled '6 blind men examining an elephant'.

1) is it tp per round
2) is it tp per damage done
3) is it tp per second
4) does it include the effects of penance
5) is it in a mass mob fight (dynamis), small group short fight (merit) or small group long fight (bcnm)
6) does the mob self generate tp

Any combo of these will give a different ranking and importance of the issue. Overall, if the melee is carrying their weight in damage, I think its generally not worth the concern.

Does it matter in solos or duos, yes, to some extent.

I just get concerned when I'm in meripo and some guy is hitting for 6-7 damage on a mob with a nasty tp move, or we are doing a long bcnm with a boss with killer moves.

[Appie]

Well back then there was no Penance, so it leaves 5 blind men? =D
The only reason I ever got involved in the argument back then was because I thought chi blasting some mobs was bullshit when I could DD it better and just as TP efficient or better as other classes (and a whole lot better than THF XD)

Back on topic: I got nothing.