Forum Settings
       
Reply To Thread

Epona's on WSsFollow

#52 May 04 2011 at 4:18 PM Rating: Excellent
***
2,236 posts
Quote:
How do I quote the parse data like pchan did?


For the tab you want to copy, have that tab up and go to the Edit menu and pick Copy as BBCode. You can then paste that directly into your post.

Note that Alla uses a bad font for preformatted text (which I've disabled via Stylish) that doesn't have the same character width when bold as its non-bold version, **** up the alignment. You may wish to delete the B tags to clean that up, if you're feeling picky.


Getting that into matlab certainly makes it easier to do the longer analysis, though my suggestion was simply to analyze the first 6-9 minutes of the data. If SD does indeed fall off without renewal, then averaging the first few minutes in with all the other intervals wipes out any possible indication of that.

As for the overall graph, I'd suggest getting ((timestamp-offset)/60) % 3) for your x coordinate to group into 1-minute blocks, and then getting the mean, count and std dev for each of those blocks.

#53 May 04 2011 at 4:39 PM Rating: Decent
**
983 posts
Oh, I did the check for that already and it doesn't seem to be the case (6 DAs in the first three minutes, 10 DAs in the second three minutes). It's difficult to say conclusively though, because I only attack about 10 times per minute.

Edit: I'll add that the only reason we believe Saber Dance works on 1 minute windows is because of a wiki post ages ago. Every other piece of information that post provided was shown to be wrong, so I guess I was just clinging onto the 1 minute window idea for simplicity's sake.

If you remove that as an assumption and look at the unsmoothed data (and accept that I was "using Saber Dance" just before the end of the graph), it looks like it starts around 50% and decays by about 5% per swing to a floor of around 15% (8 steps to the floor), similar to the way Fan Dance works (100%->20%, 8 steps to the floor). I am going to try and feed Saber Dance activation timestamps from my log to the program and see if I can align to them and simply look at probability per round after that, but a friend just proposed so I need to go out for drinks first!

Edited, May 4th 2011 6:56pm by Byrthnoth

Edited, May 4th 2011 7:02pm by Byrthnoth
____________________________
Yay for Jhereg!
http://www.ffxiah.com/player/Lakshmi/Byrth
#54 May 04 2011 at 7:00 PM Rating: Good
**
415 posts
I'm going to assume that the DA rate is linearly decreasing between 0 and 5 minutes. Without going into the math it's quite obvious that

* If you use saber dance every 5 minutes you average DA rate is 1/2 of it's inital value ( which I will call DA)
* If it overwrites itself then the DA rate is periodic over time and lineraly decresing from DA to 3/5 DA every 3 minutes so It averages to 4/5 DA.

So If saber dance overwrites itself 4/5 DA= 18.65 +/- 1.16 => 21.86% < DA < 24.76%

* If saber dance doesn't overwrite itself the da rate is still periodic : linearily decreasing from 0 to 5 minutes between DA and 0, and 0 between 5 and 6 minutes, and the average DA rate would be DA*5/12 (which is the average of the above function on its period).

In this case DA*5/12= 18.65 +/- 1.16 => 41.97% < DA < 47.54%

Before going any further, was your latest test done with zero DA gear ? If so it kinda shows that samba overwrites itself ( 50% initial DA would be noticable lol). To know accurately the initial DA value you just have to use saber dance every 5' instead of 3' and multiply the DA on the parse by 2.

Edited, May 4th 2011 9:01pm by lynnminmay

Edited, May 4th 2011 9:03pm by lynnminmay
#55 May 04 2011 at 8:26 PM Rating: Decent
**
983 posts
It seems that DA probably does decrease with Saber Dance, and probably linear (piecewise). I extracted Saber Dance timestamps in matlab and used them to separate on a first-swing, second-swing, etc. basis. This is what I came up with:
Picture

Now! The weird part is that I also went back and verified where I used Saber Dance on the previous graphs. It was 4 seconds before my first swing (the offset), so the 50% spike is probably real. I don't know if SE messed up the timescale or what, but it looks like it decays very very quickly (within the first 5-8 hits) to 15% where it pretty much stays.

I'm just leaving my parse to run overnight. If I remember stats correctly, doubling sample size decreases variance by root(2), but it's something. The current data set I'm running on is 7.5 hours, so 10 more hours should be worthwhile.

PS. Oh yeah, I was wearing 0 DA gear for the most recent data/graphs. Sorry. I'm a few drinks heavier.

PPS. I'm so **** This **** looks like biological data. Why the **** do I play a game if I have to deal with the same **** as at work? ANGRYFACE.

Edited, May 4th 2011 10:32pm by Byrthnoth
____________________________
Yay for Jhereg!
http://www.ffxiah.com/player/Lakshmi/Byrth
#56 May 05 2011 at 6:58 AM Rating: Decent
**
983 posts
Okay, aggregate statistics:

Box step: (62.67% Land rate +/- 1.32%)
used - 5192 times
Landed - 3254 times

Hit rate: (58.60% Hit rate +/- 0.73%)
Hits - 10138
Misses - 7162

Crits - 1353 (13.35% +/- 0.66%)
Extra attacks - 2652 (18.10% +/- 0.62%)

I realize that it would be better for me to have used a higher or lower Hit rate in terms of statistical power, but I was hoping to turn this into a test of "Enhances Step Accuracy" potency (3 different items) so I was saving my statistical power-ups for there. I'd always assumed there was an innate hit rate penalty on steps though, so I'm not sure how much I actually care about the potency of these items anymore. I might test Feather Step's crit rate boost first/at the same time.

I will get an Acerbic Sash and make myself a Lady Bell for my next tests, I think, as it should speed up data collection dramatically.

Edit: Oh yes, all of these tests were done on the same Hpmede (>315 DEF, so I assumed highest level).

Edit2: Less variance in Saber Dance use times this time. The unsmoothed version looks pretty striking <_<

What worries me more than anything else is that the spike decays over about 10 seconds. The highest point in the spike are oddities (low count, 1 and 5 for the first two points), but the rest are pretty high sample size.

So theories for how Saber Dance might work:
1) DA rate decays in 1 minute windows. - Not supported by the data
2) DA rate decays a lot faster than that (~10 seconds) - Somewhat supported by the most recent data.
3) DA rate decays with every swing, regardless of whether DA procs - Doesn't appear to, but maybe confounded by overlapping attack rounds
4) DA rate decays whenever a DA procs - Haven't even written the code for this yet.

Edit3, question for Kinematics:
Minor issue, but KParser determines groupings of attacks for the "Extra attacks" analysis and provides the average attack time for each group, correct? It seems much more likely to me that (if it were using a DA rate decaying with time) the DA rate would be determined slightly previous to the first attack of each round. Using the average round pushes back the timestamp and makes weird things like the 2 point in the unsmoothed graph above. Is there a way to do this without writing a grouping algorithm myself?

Edit4: I prefer this version
I aligned it to Saber Dance onset, but it's still relative to time. It is difficult for me to tell the difference between a fast decaying mechanism and one of the other options above, because they all come out looking about the same due to my relatively constant attack speed. When I make a Lady Bell and run the test I should know for sure.

Edit5: Summing everything after "50" where it looks like the crit rate has gone to baseline gives a crit rate of 16.75%. An odd number to be sure. Close to 1/6, but otherwise unremarkable!

Edited, May 5th 2011 10:45am by Byrthnoth
____________________________
Yay for Jhereg!
http://www.ffxiah.com/player/Lakshmi/Byrth
#57 May 05 2011 at 10:06 AM Rating: Excellent
***
2,236 posts
Quote:
Minor issue, but KParser determines groupings of attacks for the "Extra attacks" analysis and provides the average attack time for each group, correct?


The timestamp KParser gives each group is the time of the first attack in each group. All other attacks are considered to have 'actually' occurred at that point, but the messages show up later.

Quote:
Is there a way to do this without writing a grouping algorithm myself?


I'm not entirely sure what you're asking for, but I -think- KParser is doing roughly what you want.

Quote:
I aligned it to Saber Dance onset, but it's still relative to time. It is difficult for me to tell the difference between a fast decaying mechanism and one of the other options above, because they all come out looking about the same due to my relatively constant attack speed. When I make a Lady Bell and run the test I should know for sure.


At the moment it's looking like it behaves much more like Fan Dance, which decays per hit you take. From your data, I'm inclined to believe that Saber Dance decays either per DA or per hit you land. If the Lady Bell test takes longer to decay then I would believe it decays per hit; otherwise, decay per DA.

Quote:
Edit5: Summing everything after "50" where it looks like the crit rate has gone to baseline gives a crit rate of 16.75%. An odd number to be sure. Close to 1/6, but otherwise unremarkable!


You mean DA rate, right? :)


As a side note, I'm running my own confirmation of the DA/TA test. Unfortunately the results are looking rather problematic, and I'm restarting the test (noticed I'd had a torrent running, which could conceivably have caused lag spikes that messed up the results, even though it usually doesn't).

Setup: thf/whm, 13% DA (Brutal+Atheling+Pole+Epona), 13% TA (trait+merits+Epona)

Parse results:
6121 rounds, 792 DA (12.94 %), 471 TA (7.69 %)

So.. yeah.
#58 May 05 2011 at 10:44 AM Rating: Decent
**
983 posts
Kinematics wrote:
The timestamp KParser gives each group is the time of the first attack in each group. All other attacks are considered to have 'actually' occurred at that point, but the messages show up later.


Okay, that answers everything. I must have just looked at a strange case, or everything is offset.

"Kinematics" wrote:
Quote:
Edit5: Summing everything after "50" where it looks like the crit rate has gone to baseline gives a crit rate of 16.75%. An odd number to be sure. Close to 1/6, but otherwise unremarkable!


You mean DA rate, right? :)


Yep, sorry.

"Kinematics" wrote:
As a side note, I'm running my own confirmation of the DA/TA test. Unfortunately the results are looking rather problematic, and I'm restarting the test (noticed I'd had a torrent running, which could conceivably have caused lag spikes that messed up the results, even though it usually doesn't).

Setup: thf/whm, 13% DA (Brutal+Atheling+Pole+Epona), 13% TA (trait+merits+Epona)

Parse results:
6121 rounds, 792 DA (12.94 %), 471 TA (7.69 %)

So.. yeah.


errr... that's interesting. How would you go about calculating the variances on those? Maybe you can show which type of system it is using the variance? Upon reflection, by the time the variances differ significantly I'd expect the means would also differ significantly.

It's interesting that now we have all the possibilities represented though.
lore - TA is checked first
pchan - Mutually exclusive
kinematics - DA is checked first, kinda

The problem I see with your data is that your TA rate is significantly lower than what you'd predict in the "DA is checked first" case, right?
.87*.13 = .1131
Assuming DA procs before TA (lowering TA sample size), you'd still predict your value to be 7.69% +/- .71%.

Could the parser be splitting TA rounds into a DA and single round? How does your number of rounds compare to the number of hours you spent parsing and your effective delay?

Edited, May 5th 2011 3:43pm by Byrthnoth
____________________________
Yay for Jhereg!
http://www.ffxiah.com/player/Lakshmi/Byrth
#59lynnminmay, Posted: May 05 2011 at 5:17 PM, Rating: Sub-Default, (Expand Post) With 15% DA and 10% TA in gear I was getting 13,38 % DA and 8,31 % TA (this was the first parse I posted), which means I'm getting more TA than you even though I have 3 less TA in gear ... First step is to say that epona is broken, second step is to say what the @#%^. I don't believe the torrent lag to affect anything. I put money on epona's not doing what it says. Kine's data pretty much says "epona's is worse than 5% DA" or "3% TA is worse than 2% DA"... facepalm .
#60 May 05 2011 at 5:41 PM Rating: Excellent
***
2,236 posts
I'm currently looking into the code for the extra attack calculations in KParser. Am hoping I can resolve the issue there, if it's buggy or unreliable.

#61 May 05 2011 at 6:22 PM Rating: Excellent
***
2,236 posts
OK, ran through the code and found that one particular threshold that was under what it needed to be by 0.003 seconds was throwing everything off. Have confirmed results with a couple other test runs, but can't guarantee how it will behave under all conditions/weapons (which isn't any different than before, really). I need to come up with a better algorithm, but it's a real pain to work out.

Anyway, 13%/13% test results:

Melee Data 
 
Player               # Melee Attacks    # Melee Rounds    Attacks/Round    # Extra Attacks 
Motenten                        8072              5900                1               2172 
 
Player               # +1 Rounds   # +2 Rounds   # +3 Rounds   # +4 Rounds    # >+4 Rounds 
Motenten                     628           772             0             0               0 
 
Player               # MultiAttack Rounds    MultiAttack %     Kills w/Min Attacks    Kills w/<Min Attacks 
Motenten                             1400          23.73 %                       0                       0 
 
 
Treat As: 
 
Multi-attacks per attack (2x/3x): 
 
Player               # Double Attacks    DA Rate    Perc. DA     # Triple Attacks    TA Rate    Perc. TA 
Motenten                          628    10.64 %     44.86 %                  772    13.08 %     55.14 %


DA: 10.64% +/- 0.79% => 9.85% - 11.43%
Predicted: 11.31%

TA: 13.08% +/- 0.86% => 12.22% - 13.94%
Predicted: 13% or 11.31%

Total multi-attack: 1400/5900 = 23.73% +/- 1.09% => 22.64% - 24.82%
Predicted: 22.61% (xor) or 24.31% (priority)


This re-validates the December test data that indicated that TA had priority over DA, and implies lynn's data is incorrect.

I'll upload a new version of KParser so that you can check the parse with the corrected threshold value (does not need to be re-parsed). If the results are more in line with the expected TA>DA theory then this can be resolved as a bogus reveal based on a flaw in KParser. If it still shows the xor results then further investigation will be necessary.



New version uploaded. Installer for 1.5.12a is in the downloads section.

Edited, May 5th 2011 7:33pm by Kinematics
#62 May 05 2011 at 7:31 PM Rating: Decent
15 posts
Thanks for testing this yourself Kine. I tried to do a test of my own, but without a THF character to use, the lack of TA options (Only the 3% Epona's gives), I couldn't refute his evidence. I asked Lynn to upload his parse, but never got a reply. I have noticed that KParser consistenly had issues showing DAs as 2 seperate 1 hit rounds, and TAs as either 3 single rounds, or a DA round with a single hit round. Both greatly inflate the Attack rounds shown. I did a parse of about 2500 attack rounds that was in reality 2-300 less. I've been testing the set bonus for Monk Empyrian Armor, and have to look through the extended details, and sync them up with logger for the same reasons of round bleeding, and figured Lynn's parse had this problem. Should these updated threshold values help with h2h parsing issues? It seems it really has problems when double kicks occur, and in general always seems more wonky with h2h than single or 2H weapons.
#63 May 05 2011 at 8:29 PM Rating: Decent
**
983 posts
Okays, how do I apply the new version to my old parses? Just install it and reopen them?

I have a 15 hour parse scheduled to end at 9AM tomorrow, same conditions (0% DA) but with a lady bell and airy buckler. At the moment, switching from Cobra Staff to Lady Bell has pushed the DA rate up to 20%, which makes me think that it's time-related decay instead of something else. I'm collecting data so much faster this way... I'll probably have something conclusive by tomorrow.
____________________________
Yay for Jhereg!
http://www.ffxiah.com/player/Lakshmi/Byrth
#64 May 05 2011 at 8:40 PM Rating: Decent
15 posts
Okay, opened up a few old parses with the new version, seems much better for DA and TA rate estimates. Old parse had DA listed as 126, TA as 2 in 1348 rounds. New has DA 208, TA 47 in 1168 rounds. In reality it should be 207DA, 47TA, in 1177 rounds, so way closer to the actual in your updated KParser. However, H2H got way out of control. I had an old parse that showed:

#+1 Rnds= 158, +2=10, +3=16, +4=21, >+4=5

Now shows as:

#+1 Rnds=45, +2=1, +3=100, +4=33, >+4=25

In actuality it should show:

#+1 Rnds=241, +2=34

Is this just a speed issue, where attack rounds are too fast? or Dual-Wield/H2H related? My attack rounds would be at aprox 206delay, or 103 per fist.



#65 May 05 2011 at 9:44 PM Rating: Excellent
***
2,236 posts
Byrthnoth wrote:
Okays, how do I apply the new version to my old parses? Just install it and reopen them?

I have a 15 hour parse scheduled to end at 9AM tomorrow, same conditions (0% DA) but with a lady bell and airy buckler. At the moment, switching from Cobra Staff to Lady Bell has pushed the DA rate up to 20%, which makes me think that it's time-related decay instead of something else. I'm collecting data so much faster this way... I'll probably have something conclusive by tomorrow.


Yes, install the new version and reopen the old parses. It doesn't change any of the old data, just adjusts how it analyzes it.

If your current parse is already active, you don't have to do anything. Let it run its course, then install the new parser after that to look at the data.

#66 May 05 2011 at 9:57 PM Rating: Excellent
***
2,236 posts
Diasetsu wrote:
Okay, opened up a few old parses with the new version, seems much better for DA and TA rate estimates. Old parse had DA listed as 126, TA as 2 in 1348 rounds. New has DA 208, TA 47 in 1168 rounds. In reality it should be 207DA, 47TA, in 1177 rounds, so way closer to the actual in your updated KParser. However, H2H got way out of control. I had an old parse that showed:

#+1 Rnds= 158, +2=10, +3=16, +4=21, >+4=5

Now shows as:

#+1 Rnds=45, +2=1, +3=100, +4=33, >+4=25

In actuality it should show:

#+1 Rnds=241, +2=34

Is this just a speed issue, where attack rounds are too fast? or Dual-Wield/H2H related? My attack rounds would be at aprox 206delay, or 103 per fist.



It's largely a speed issue. There are a few different heuristics that it uses to determine if a new round has started, but it's also limited by certain thresholds, which are extremely touchy. I need to look into other new ways to determine which hits should be grouped together into a round.

If anyone feels like trying to come up with an idea, a few points:
1) Lines are grabbed at 1/2 second intervals (usually 0.5 seconds, but typically a smidge higher such as 0.501 or some such).
2) The game usually displays successive hits approximately 1 second apart (aside from bursts of attack rounds after a bout of lag), though occasionally they'll be captured 1.5 seconds apart, or slightly higher.
3) From that you can see that a triple attack round will have lines at timepoints 0.0, 1.0 and 2.0 seconds (roughly). A weapon with a delay of 180 (3 seconds) would hit its next round at 3.0 seconds, making it difficult to separate from earlier hits (new round, or quad attack proc?). Effectively the same issue with H2H.
4) I can collect time clusters of the differences in time between each hit, where you can see peaks of the most common times between hits, along with surrounding differences that are nearly the same.
5) I can do a "group by" clustering using adjacent hits with a variety of conditions in order to select into the group. This is not restricted to only 2 adjacent hits.
6) At the moment I'm doing strict thresholding. I'm mainly looking for ideas on a 'fuzzy' thresholding.
#67lynnminmay, Posted: May 06 2011 at 3:28 AM, Rating: Sub-Default, (Expand Post) With the new version : DA15 TA 10
#68 May 06 2011 at 3:53 AM Rating: Decent
**
415 posts
Next time I'll try not to interrupt the parse to check partial data. I added the total fight lengths in seconds of the different "fights" (I paused 5 times) and divided by my delay of 366 (signet staff) x 0.85 (15% haste) /60 and found 8317 rounds for the 15DA/10TA parse, so we are ~300 rounds off. I don't think lag can be the issue since you would DC after 30 sec at most.

Edited, May 6th 2011 5:54am by lynnminmay
#69 May 06 2011 at 3:59 AM Rating: Decent
**
983 posts
It looks like you're largely bound by the game display latency. Your parser just "parses" information from the chat log, correct?

It would be worthless for us here, hitting for 0s like the pros we are, but damage for the whole TA/DA is assigned when the first hit is displayed. You could use the number of times mob HPP changes to determine the number of attack rounds and attack round onset.

In general, I think you're hitting the limit of what's possible. It has been my experience that the times where kparser can't tell whether something was a DA, TA, or 4+ round, I also can't tell looking purely at the log. All the tricky programming techniques in the world aren't going to get around that.

PS. My graphics card decided to throw in the towel last night, so FFXI aborted itself. Still got 8000+ rounds out of it, averaging out to 20.5% DA. It looks like it decays over the same period (about 30 seconds from onset), but . . . it decays to a higher value somehow (~19% instead of ~17%). Theories 3 and 4 would have predicted a lower average rate and all the theories predict an unchanging floor. It looks like DA floor from Saber Dance might depend on Delay?
Either that, or it depends on something that's correlated for each Saber Dance use but decorrelates pretty quickly. I have lower hit rate with lady bell than I did with staff. The most annoying case would be case 4 with the additional stipulation that the DA proc has to land. We shall see!


Edit:
pchan, some people that do a lot of delay testing have proposed that it's actually closer to ~58 delay per second. They were more exact than that, but I can't remember exactly what they said.

Edited, May 6th 2011 6:01am by Byrthnoth
____________________________
Yay for Jhereg!
http://www.ffxiah.com/player/Lakshmi/Byrth
#70lynnminmay, Posted: May 06 2011 at 4:33 AM, Rating: Sub-Default, (Expand Post) Ah well if it's 58 delay the number of rounds seems correct at least which leaves us with another problem : TA rate is too borderline to validate anything. I guess I will run the same tests again with zero% haste and without interruption and I will check that both parse give the same delay.
#71 May 06 2011 at 5:20 AM Rating: Good
*
52 posts
I remember VZX back in old times translated from StudioGobli something like 1sec = 58.75 delay ? (but as usual 0 test data...)

Maybe slightly offtopic, but i discovered very weird parsing of DArate from KParser while testing Ravager set bonus. (33% DA equipped but ~17% observed, decent sample size)

I have also made 5 tests on Fortifications to "verify" KParser's results compared to RAWdata tab imported into Excel => own parsing(+ manual counting) using the timestamps. It shows quite some divergences...
i'm wondering if KParser can use the PC clock instead of timestamps for more precision at parsing multiattacks? Not sure but maybe PC clock has "more" millisecs precision than timestamps limited to :xx ?

I'll post those 5 tests in same issue tag already posted 2 months ago in KParser Issues page.

Edited, May 6th 2011 8:30am by Masamunai
#72 May 06 2011 at 5:28 AM Rating: Decent
**
415 posts
I just thought of this idea for checking the accuracy of the rounds count : I'll parse my self together with my mule on a low level job with a signet staff equipped, while wearing no haste gear. It seems to be a 100% safe way to have the exact value of the average #hits/round since the parse of my char will give the exact number of hits landed while the parse of my mule will give the exact number of rounds (assuming of course that I have them engage at the same time and stop at the same time altough it should resuslt in +/- 2 rounds at worse).
#73 May 06 2011 at 6:38 AM Rating: Decent
**
415 posts
As far as the alogorithm for finding DA and TA rates goes ... Would it not be simpler to only identify the rounds without extra attack (seems simple as they are the majority and they are the rounds with 2 attacks clearly sperated by > 2 sec) ? Then by a delay calculus ( the delay is the majority of the delay between 2 attacks observed, and not the maximum delay, to avoid lag), extrapolate the global amount of rounds (since delay is constant and not a random variable it should be exactly the real number of rounds unless you have 100% extra rate or something dumb).

Call N_1= total number of rounds, N_2= total number of single rounds and N the number of attacks.
If DA is the effective DA rate and TA is the effective TA rate (effective means what you see in the game not in the stats) then
N/N_1= 1+ DA+2*TA
(N_1-N_2)/N_1= DA+TA
Which you can solve easily :
DA=-(N-3*N_1+2*N_2)/N_1 and TA=(-2*N_1+N_2+N)/N_1

I think you will be losing the confident interval though if you don't consider checking the amount of attack per individual round. But at least it gives the exact DA rate and TA rate in your data. (After thinking about it you should actualy be able to find a CI since you know excatly the amount of DA and the amount of TA that occured).

For instance : If I find, with the algorithm,
N_1=10,000 rounds with my algorithm
N_2=8,000 rounds without extra attacks
N= 13,000 swings

Then NECESSARILY I had exactly 1000 tripple rounds and 1000 double rounds, which allows me to compute the confident intervalls...


EDIT : If you use my data for the DA15/TA10 parse and you assume that Kparser is counting the rounds correctly, then you have
N_2:=8655-1161-812-4-3-2;N_1:=8655;N:=11474;
Which gives using the above formula
da=13.23% +/- 0.71 and ta=9.67% +/- 0.62%
ta : [9.05,10.29]
da : [12.49,13.95]

Now this time it is really matching TA being checked first and excluding DA first and xor.

Now for the parse with 17DA/10TA
N_2:=8464-1323-802-9-3-1;N_1:=8464;N:=11435;
Using the above formula
da=15.41 +/- 0.77 and ta=9.84 +/- 0.63
da : [14.64,16.18]
ta : [9.21, 10.47]

This time it is also matching TA checked first and excluding DA first and xor.



Edited, May 6th 2011 9:13am by lynnminmay
#74 May 06 2011 at 7:38 AM Rating: Decent
**
983 posts
Doh ; ; haha. The new parser version tells me my double attacks are triples. If we don't suspect the old version was having issues with double attacks, maybe I'll just keep using it.


1113 DA/3903 rounds (+8% DA) went to 1071 DA rounds with 16 +2 rounds and 41 +3 rounds out of 3790 rounds total.

3790 + 42*2 + 16*2 = 3904, but that implies that the +3 rounds were previously interpreted as one DA and two single hits, and the +2 rounds were previously interpretted as three single hits.
____________________________
Yay for Jhereg!
http://www.ffxiah.com/player/Lakshmi/Byrth
#75 May 06 2011 at 7:52 AM Rating: Good
*
52 posts
Rounds count isnot a prob for KParser: my own parsing on Excel gets very close results if not exact same numbers.

It's the parsing of multiattacks that pose problem...

From that discovery you can guess the repercussion on the validity of all tests done in this topic regarding DA/TA...
#76 May 06 2011 at 8:00 AM Rating: Decent
**
983 posts
Well, the number and type of multi-attack are entirely determined by when it chooses "rounds" to begin. I can provide the problematic data if anyone is interested in taking a look.

Gear was:
Lady bell/Airy Buckler/______/________
AF3+2/_______/_______/Brutal
AF3+2/AF3+2/Hoard/Malflame
Atheling/Taru RSE waist/AF3+2/AF3+2
16% Haste, +8% DA, Saber Dance used every 181 seconds. Box Step ~10 times per cycle and Reverse Flourish ~5 times per cycle.

I can see why it was having trouble, 216*.84 = 181.44, so I'm very close to 3 seconds between rounds.
____________________________
Yay for Jhereg!
http://www.ffxiah.com/player/Lakshmi/Byrth
#77 May 06 2011 at 8:51 AM Rating: Decent
**
415 posts
stupid AF3+2 and their haste lol.
#78 May 06 2011 at 9:18 AM Rating: Decent
**
983 posts
Haha, well if there was a way to force it to only recognize double attacks it would all be fine.
____________________________
Yay for Jhereg!
http://www.ffxiah.com/player/Lakshmi/Byrth
#79 May 06 2011 at 1:09 PM Rating: Excellent
***
2,236 posts
masamunai wrote:
i'm wondering if KParser can use the PC clock instead of timestamps for more precision at parsing multiattacks? Not sure but maybe PC clock has "more" millisecs precision than timestamps limited to :xx ?


The timestamps are themselves generated from the PC's clock. It does not use the "Timestamp" plugin (if that's what you were suggesting), except in cases where you're parsing from log files rather than RAM, and the plugin's values are present.

It has a resolution of 0.5 seconds to make sure it gets frequent data, but not so fast as to hammer the user's machine (primarily a concern with older, slower computers), and minimize the impact of running the parser alongside the game.


On the delay, I believe VZX put it at 58.85 delay per second, which is about the same as what I calculated it to be at one point. However it's an annoying number to use, and everyone is pretty comfortable with 60/second. I suppose we should consider the more precise value for certain types of testing.

Quote:
I just thought of this idea for checking the accuracy of the rounds count : I'll parse my self together with my mule on a low level job with a signet staff equipped, while wearing no haste gear. It seems to be a 100% safe way to have the exact value of the average #hits/round since the parse of my char will give the exact number of hits landed while the parse of my mule will give the exact number of rounds (assuming of course that I have them engage at the same time and stop at the same time altough it should resuslt in +/- 2 rounds at worse).


Nice idea. Though may be simpler to have them already engaged when you start the parse, and stop the parse before disengaging. That should get you +/- 1 round.

Quote:
I can see why it was having trouble, 216*.84 = 181.44, so I'm very close to 3 seconds between rounds.


And that's why I told you 200-210 delay minimum :p
#80 May 06 2011 at 1:14 PM Rating: Excellent
***
2,236 posts
Thoughts on the extra attacks formulations:


I have N individual hits. > list(N)

From list(N) I can pull the timestamps for each hit > list(T)

From list(T) I can generate list(dT) of the difference in timestamps between hits. This will have at most N-1 entries.

I can then collect frequency data on the dT values, for an array of freq(dT).

I should then generally have multiple peaks within freq(dT):

Ex: 366 delay ~= 6 seconds for the signet staff

lowest peak: difference between two successive hits in a multi-hit round (~1 second)
highest peak: difference between two non-multi-hit rounds, or last hit of a 2-hit round (DW or H2H) and first hit of the next.
middle peaks: difference between the end of a multi-hit round and the start of the next round.

Example peaks:

1 s = first and second hit of DA round, or second and third hit of a TA
4 s = last of triple round and start of next round
5 s = last of double round and start of next round
6 s = first of one-hit round and start of next round

So what we want is the highest peak that doesn't involve changing mobs (those diffs should be excluded from list(dT)).

This peak should account for at least 20% of all attacks (worst case: testing OA2-4 weapon on Footwork, with 75%-80% DA rate).

Take the low-end value of the peak (ie: surrounding variance) as typical margin of time between attack rounds. We then want to group together all attacks that occurred at less than that minimum.

The grouped attacks should have dT values within the range of the first peak.

It should group hits that are within range of < high peak. It should not group hits that are outside the range of the low peak (given some buffer, making that 2 seconds).


All well and good, and putting together an algorithm that works for this specific construct is easy. The problem arises when trying to also deal with possible dual-wield or H2H, where you have a minimum of 2 hits per round.

In that case your high peak difference will never be the actual round duration, but generally 1 second less. Using the same algorithm here creates a very high risk of cutting off high multi-hits from a round.

What we want to look for is if there is always (90%+) an adjacent hit within the minimum peak difference. If so, that should be a solid indicator that we're attacking with base 2 hits per round.

So if I put the hits per round calculation in -before- trying to group rounds instead of after, I think it will work out...


I've readjusted the code to account for the above considerations. Uploading as 1.5.12b so you can check the results of any parses you have. I've run through several different parses, and it appears to be working ok. There are still some invalid counts, but it seems to handle the vast majority of the multi-hit rounds properly even in difficult parses, and comes out near perfect in well-designed tests.
#81 May 06 2011 at 1:28 PM Rating: Good
**
983 posts
These version changes are having a dramatic effect on my results. Are we confident that the most recent version is the most accurate? It has made my results much more consistent, at least.


With the newest new version, I'm at N=3844 melee rounds and 30.5% DA rate with DA+8% (Lady bell, Atheling and Brutal) on DNC/WHM.
Almost exactly 22.08% DA with Cobra Staff on DNC/WHM (N~10000).
Almost exactly 21.88% DA with Lady bell on DNC/WHM with no DA (N=8785) <--- 20% DA rate after/before the spike

I'd like to sit down and do the further analysis on this, but it must wait!


Did a little analysis while at work on the biggest data set. Very nice and definite spike dropping from .5 to .2 over the first 30 seconds after JA usage. Assuming the drop is linear, this means that you drop 1% DA per second for 30 seconds after JA usage, which would predict a 22.5% average usage (in fact it's lower in my script, because I spam two JAs in a row after Saber Dance... I should consider changing that xD.) The attack rounds after JA usage analysis didn't look nearly this nice, so it's probably related to time.

I'm going to wait until I've done the rest of the analysis to draw any major conclusions (and change my merits), but if this is the mechanism then 5/5 Saber Dance vs. 1/5 nets out to a 1% increase in Double Attack rate.

Here is the Cobra Staff plot, and I linked one for the Lady Bell above in this post. It looks pretty consistent with respect to time.

Unless anyone else has something to add, I'm fairly confident in this mechanism and will focus my attention more on steps now (though still collect data for Saber Dance, of course).

Edited, May 6th 2011 4:46pm by Byrthnoth
____________________________
Yay for Jhereg!
http://www.ffxiah.com/player/Lakshmi/Byrth
#82 May 06 2011 at 1:47 PM Rating: Decent
**
415 posts
Awesome work kinematics i'll try this in a few hours during my sleep..
#83 May 06 2011 at 2:21 PM Rating: Excellent
***
2,236 posts
Well, I'm fairly confident it's a better fit :)

I'd like to hear confirmation from lynn on the DA/TA test, Diasetsu on his h2h parse, and Masamunai on the Ravager parse before putting a full stamp of approval on it, though.

Actually, Masa already uploaded copies of his parses, so I can look at them now.

Parse 1, 4 pieces of Ravager gear.
Expected parse rate: unknown
Manually parsed rate (from Masa): 35%

Melee Data 
 
Player               # Melee Attacks    # Melee Rounds    Attacks/Round    # Extra Attacks 
Masamunai                       1263               954                1                309 
 
Player               # +1 Rounds   # +2 Rounds   # +3 Rounds   # +4 Rounds    # >+4 Rounds 
Masamunai                    309             0             0             0               0 
 
Player               # MultiAttack Rounds    MultiAttack %     Kills w/Min Attacks    Kills w/<Min Attacks 
Masamunai                             309          34.60 %                      61                       0 
 
 
Treat As: 
 
Multi-attacks per attack (2x/3x): 
 
Player               # Double Attacks    DA Rate    Perc. DA     # Triple Attacks    TA Rate    Perc. TA 
Masamunai                         309    34.60 %    100.00 %                    0     0.00 %      0.00 %


So that looks like it's working well.

Second parse:

Melee Data 
 
Player               # Melee Attacks    # Melee Rounds    Attacks/Round    # Extra Attacks 
Masamunai                       1294               990                1                304 
 
Player               # +1 Rounds   # +2 Rounds   # +3 Rounds   # +4 Rounds    # >+4 Rounds 
Masamunai                    304             0             0             0               0 
 
Player               # MultiAttack Rounds    MultiAttack %     Kills w/Min Attacks    Kills w/<Min Attacks 
Masamunai                             304          33.30 %                      77                       0 
 
 
Treat As: 
 
Multi-attacks per attack (2x/3x): 
 
Player               # Double Attacks    DA Rate    Perc. DA     # Triple Attacks    TA Rate    Perc. TA 
Masamunai                         304    33.30 %    100.00 %                    0     0.00 %      0.00 %


Looks like similar amounts. Neither parse has any spurious attack rounds (2 or more extra attacks). Overall, quite happy with it.



Edited, May 6th 2011 3:25pm by Kinematics
#84lynnminmay, Posted: May 06 2011 at 4:58 PM, Rating: Sub-Default, (Expand Post) It's finding 10 and 26 +3 rounds for me >_>. So here is a link if you want to dissecate it
#85 May 06 2011 at 4:59 PM Rating: Decent
Scholar
***
1,423 posts
Any way I can aid in this testing at all?

I haven't played solidly (logged in a few times to say hi to a few ppl and help a rl friend w/ Epona's ring) in about a month, but I'll hop on to help in testing, provided there's actually something I can do that'll be worthwhile.
#86 May 06 2011 at 6:25 PM Rating: Decent
15 posts
Newest Version appears to be pretty **** good. Almost all my H2H parses have the attack rounds within 2 rounds of what I recorded from correcting with logger. It does still seem to push some 4hit rounds into 2 3 hit rounds, but even that doesn't skew results all that much. I could still verify future parses with logger, but if the difference ends up being fairly insignificant I probably won't. Either way where I would have had to of spent 30mins correcting a parse, with the new rules for attack rounds you changed, it would probably only take about 5mins to adjust. I've pretty much only been logging in on the weekends to do fortification testing for Empyrian Armor, but if I manage to get a chance to do some testing at 80% speed reduction, I'd love to see how it handles super fast attack speeds. Also out of curiosity, how does KParser distinguish between a 3hit round with 2attacks plus a DA, and a round with 2attacks plus a KA?
#87 May 06 2011 at 6:45 PM Rating: Decent
15 posts
Veggeto wrote:
Any way I can aid in this testing at all?

I haven't played solidly (logged in a few times to say hi to a few ppl and help a rl friend w/ Epona's ring) in about a month, but I'll hop on to help in testing, provided there's actually something I can do that'll be worthwhile.



If you could help add data to my AF3 full set bonus it would be much appreciated. I haven't really had the time to get as nice of a CI as I'd like for the data I have. Off the top of my head I think it's still +/- 1.5% or somewhere around there.
#88 May 07 2011 at 1:57 AM Rating: Good
*
52 posts
Excellent fix Motenten: all my war AF3 testings now matches (within confidence intervall) the equipped DA%.

Btw anyone here would have discovered that if done a control test before throwing asap on DA/TA tests.

Also, i'm interested in answers for Diasetsu's questions at the end of his post.

Edited, May 7th 2011 4:04am by Masamunai
#89 May 07 2011 at 4:49 AM Rating: Good
*
52 posts
Actually, after further verification on old parses i had, with KP 1.5.12b, i found one showing divergences:

Player Info: WAR75/nin37 Suppano HasteGears16% StoreTP+7 Maneater/Rapier STR82+20 DEX66+30 Att422 BrutalEar
Mob = Fortification

So guessing DA10% + 4%merits + 5% Brutal = 19% DA expected.

Offense tab results :

Damage Summary 
Player               Total Dmg   Damage %   Melee Dmg   Range Dmg   Abil. Dmg  WSkill Dmg   Spell Dmg  Other Dmg 
Masamunai                28274    99,95 %       17807           0           0       10467           0          0 
SC: Scission                13     0,05 %           0           0           0           0           0          0 
Total                    28287   100,00 %       17807           0           0       10467           0          0 
 
Melee Damage 
Player            Melee Dmg   Melee %   Hit/Miss    M.HR %   M.Acc %  M.Low/Hi  M+0.Avg  M-0.Avg  #Crit  C.Low/Hi   C-0.Avg     Crit% 
Masamunai             17807   62,98 %    1546/72   95,55 %   95,55 %      0/15     9,84     9,84    361      0/23     17,02   23,35 % 
 
Weaponskill Damage 
Player                 WSkill Dmg   WSkill %  Hit/Miss   WS.Acc %   WS.Low/Hi   WS.Avg 
Masamunai                   10467    37,02 %      95/0   100,00 %      64/157   110,18 
 - Rampage                  10467   100,00 %      95/0   100,00 %      64/157   110,18 
 
Skillchain Damage 
Skillchain          SC Dmg  # SC  SC.Low/Hi  SC.Avg 
SC: Scission            13     2        5/8    6,50


Extra Attacks tab results:

Melee Data 
 
Player               # Melee Attacks    # Melee Rounds    Attacks/Round    # Extra Attacks 
Masamunai                       1618               677                2                264 
 
Player               # +1 Rounds   # +2 Rounds   # +3 Rounds   # +4 Rounds    # >+4 Rounds 
Masamunai                    214            25             0             0               0 
 
Player               # MultiAttack Rounds    MultiAttack %     Kills w/Min Attacks    Kills w/<Min Attacks 
Masamunai                             239          35,30 %                       0                       0 
 
Treat As: 
 
Multi-attacks per attack (2x/3x): 
 
Player               # Double Attacks    DA Rate    Perc. DA     # Triple Attacks    TA Rate    Perc. TA 
Masamunai                         264    19,50 %    100,00 %                    0     0,00 %      0,00 % 
 
Multi-attacks per round (Kicks): 
 
Player               Footwork?    # Rounds w/Kicks    Kick Attack Rate 
Masamunai                   No                 214             32,82 %


Multi Attacks% = 35.50% <---- HUH?!
DA Rate = 19.50% <---- Allright !

Now another old parse, same as the above one but replaced Rapier offhanded with Joyeuse :

Offense tab results :

Damage Summary 
Player               Total Dmg   Damage %   Melee Dmg   Range Dmg   Abil. Dmg  WSkill Dmg   Spell Dmg  Other Dmg 
Masamunai                26462    99,98 %       16527           0           0        9935           0          0 
SC: Scission                 5     0,02 %           0           0           0           0           0          0 
Total                    26467   100,00 %       16527           0           0        9935           0          0 
 
Melee Damage 
Player            Melee Dmg   Melee %   Hit/Miss    M.HR %   M.Acc %  M.Low/Hi  M+0.Avg  M-0.Avg  #Crit  C.Low/Hi   C-0.Avg     Crit% 
Masamunai             16527   62,46 %    1456/66   95,66 %   95,66 %      0/16     9,70     9,70    345      0/23     16,67   23,70 % 
 
Weaponskill Damage 
Player                 WSkill Dmg   WSkill %  Hit/Miss   WS.Acc %   WS.Low/Hi   WS.Avg 
Masamunai                    9935    37,54 %      90/0   100,00 %      68/162   110,39 
 - Rampage                   9935   100,00 %      90/0   100,00 %      68/162   110,39 
 
Skillchain Damage 
Skillchain          SC Dmg  # SC  SC.Low/Hi  SC.Avg 
SC: Scission             5     1        5/5    5,00


Extra Attacks tab results:

Melee Data 
 
Player               # Melee Attacks    # Melee Rounds    Attacks/Round    # Extra Attacks 
Masamunai                       1522               931                2                  3 
 
Player               # +1 Rounds   # +2 Rounds   # +3 Rounds   # +4 Rounds    # >+4 Rounds 
Masamunai                      3             0             0             0               0 
 
Player               # MultiAttack Rounds    MultiAttack %     Kills w/Min Attacks    Kills w/<Min Attacks 
Masamunai                               3           0,32 %                       0                       0 
 
Treat As: 
 
Multi-attacks per attack (2x/3x): 
 
Player               # Double Attacks    DA Rate    Perc. DA     # Triple Attacks    TA Rate    Perc. TA 
Masamunai                           3     0,16 %    100,00 %                    0     0,00 %      0,00 % 
 
Multi-attacks per round (Kicks): 
 
Player               Footwork?    # Rounds w/Kicks    Kick Attack Rate 
Masamunai                   No                   3              0,32 %


Less than 1% DA o.O, surprising isnot it ?

Posted parses on KP site @ Issue81.

Edited, May 7th 2011 8:03am by Masamunai
#90 May 07 2011 at 6:02 AM Rating: Decent
**
983 posts
The first parse is just the result of dual wield:
1618/(677*2) = 1.19, with ~35% of the rounds containing at least one DA. (.19*2-.19*.19 = 34%, so about right)

The second parse... I'm less sure!
____________________________
Yay for Jhereg!
http://www.ffxiah.com/player/Lakshmi/Byrth
#91 May 07 2011 at 9:33 AM Rating: Decent
**
415 posts
I parsed myself 16 hours straight without pause, together with my mule. Both had signet staff,0% haste, my mule had no DA or TA gear or trait. I myself was THF/WAR level 76 with brutal and max TA merits so this is 15% DA and 10% TA.

Melee Data 
 
Player               # Melee Attacks    # Melee Rounds    Attacks/Round    # Extra Attacks 
Mulasse                         9448              9448                1                  0 
Pimpchan                       12621              9447                1               3174 
 
Player               # +1 Rounds   # +2 Rounds   # +3 Rounds   # +4 Rounds    # >+4 Rounds 
Mulasse                        0             0             0             0               0 
Pimpchan                    1261           953             1             1               0 
 
Player               # MultiAttack Rounds    MultiAttack %     Kills w/Min Attacks    Kills w/<Min Attacks 
Mulasse                                 0           0,00 %                       0                       0 
Pimpchan                             2216          23,46 %                       0                       0 
 
 
Treat As: 
 
Multi-attacks per attack (2x/3x): 
 
Player               # Double Attacks    DA Rate    Perc. DA     # Triple Attacks    TA Rate    Perc. TA 
Pimpchan                         1261    13,35 %     56,96 %                  953    10,09 %     43,04 % 
 
Multi-attacks per round (Kicks): 
 
Player               Footwork?    # Rounds w/Kicks    Kick Attack Rate 
Pimpchan                   Yes                1261             14,85 % 
 
Multi-attacks per attack (Zanshin): 
 
Player               # Missed First Attacks    # DA w/Missed First   Possible Zanshin % 
Pimpchan                               7581                   1027              13,55 % 
 
Player               Acc. Rate of First Attacks    Acc. Rate of Second Attacks 
Pimpchan                                19,75 %                        20,35 % 
 
 
 


* same number of rounds on both chars => It means Kparser correctly distinguishes between DA/TA rounds and simple rounds
* Me and my mule have the same number of rounds +/- 1. (9447 or 8).
* TA rate : 10.09% +/- 0.60 = > looks good
* DA rate : 13.35% +/- 0.68 => looks good to be 0.90*0.15=13.5

As I explained earlier you can double verify DA rate and TA rate since you know excatly the amount of rounds, thanks to the mule. 9448 rounds but 12621 attacks means 1.336 average attacks per rounds. This matches TA being checked first since
1+2*0.10+0.15*0.90=1.3350 (no confident interval though, just law of large numbers, N=9448 is big yo)





edit : I've wanted to check impetus with this method since it was released. Just need enough - str on mnk using a dmg1 club to avoid damaging. I calculated that this gear would work on my lol elvaan **** sub ranger 1 with max merits (not like str matters ANYWAY lol abyssea):

http://www.ffxiah.com/members/item-sets?id=200282
You would need to automate impetus every 3 minutes and check the avergae crit rate to see if it matches +1% per successful hit. For safety you can spam impetus every 1 second. Thanks to level 90 mnk should cap accuracy at 90 with a club...



Edited, May 7th 2011 11:52am by lynnminmay
#92 May 07 2011 at 2:39 PM Rating: Excellent
***
2,236 posts
Masamunai wrote:
Btw anyone here would have discovered that if done a control test before throwing asap on DA/TA tests.


Personal embarressment is a wonderful motivator to actually fix stuff :)

Diasetsu wrote:
Also out of curiosity, how does KParser distinguish between a 3hit round with 2attacks plus a DA, and a round with 2attacks plus a KA?


It doesn't. It's up to you to set up the conditions so that you know what you're getting. The restrictions on the KA percentages, though, is that you're expected to have no DA/TA/etc gear at all, and test only for KAs, so it only counts 3-hit rounds as valid for rate purposes (or 2 hit rounds if using Footwork).

Masamunai wrote:
Multi Attacks% = 35.50% <---- HUH?!
DA Rate = 19.50% <---- Allright !


Since you can DA on both hands in that setup, Multi Attacks includes all DAs on both hands, but on a per-round basis. So yeah, it'll be about twice the DA rate.

#93 May 07 2011 at 9:08 PM Rating: Decent
**
983 posts
Well, I did the necessary control with a 5 minute SD rotation. It's pretty much the same mechanism we've been seeing.

20.96% (+/-.88%) DA 1702/8121

I still do SD -> NFR -> Box Step, so the DA spike is underrepresented in my global average because it decays so quickly.

Tonight I'm going to test how Saber Dance stacks with /WAR (finally got a damned Sash). Tomorrow night I'll test what AF3+2 and 5/5 Feather Step do. After that, it's the update and I probably won't be doing much testing for a while.
____________________________
Yay for Jhereg!
http://www.ffxiah.com/player/Lakshmi/Byrth
#94 May 08 2011 at 4:03 AM Rating: Decent
**
415 posts
I parsed the whole night with lady bell, on monk, spaming impetus every 1 second to be certain I don't miss any.

So far I average 16.39% critrate (10k hits, 95% accuracy)


Against Om'hpende with 62 str and 62 dex so I hit for 0 with a D1 club and I have a base 9% crit (crit rate floored + merits).

The following program simulates an infinite duration impetus and calculates the global crit rate bonus over time
 
x:=0;y:=0.0;impetus:=0;c:=0;hit:=0;N:=100000; 
> for n from 1 to N do 
> x:=stats[random,uniform[0,1]](2); 
> if x[1]<=0.95 then hit:=hit+1; end if; 
> if (x[2]<=c/100) and (x[1]<=0.95) then y:=y+1; end if: 
> if x[1]<=0.95 then c:=min(c+1,50); else c:=0; end if: 
> end do: y/hit; 


with 100k rounds we find 17.4% on average

The following program simulates the use of impetus every 5 minutes and calculates the global crit rate bonus over time (203 delay = what I get with lady bell and 6% haste, and crit bonus capping at 50%)

 
restart;delta:=203;x:=0;y:=0.0;impetus:=0;c:=0;hit:=0;N:=100000; 
> for n from 1 to N do 
> x:=stats[random,uniform[0,1]](2); 
> if x[1]<=0.95 then hit:=hit+1; end if; 
> if (x[2]<=c/100) and (x[1]<=0.95) then y:=y+1; end if: 
>  
> if impetus<=3*60*59 then  
>   if x[1]<0.95 then c:=min(c+1,50); else c:=0; end if: 
> else c:=0; end if: 
>  
> impetus:=impetus+delta: 
> if impetus>5*60*59 then impetus:=0; end if: 
> end do: y/hit;                               


With 100k rounds we find 7.5%

7.5% + 9%=16.5% so I think it's correct that impetus is +1% per crit. Unsure about the actual cap value since using 30% or 70% in the simulation doesn't give much difference.


I will post the final parse later.


 
 
Melee Damage 
Player            Melee Dmg   Melee %   Hit/Miss    M.HR %   M.Acc %  M.Low/Hi  M+0.Avg  M-0.Avg  #Crit  C.Low/Hi   C-0.Avg     Crit% 
Pimpchan                  0    0.00 %  15483/802   95.08 %   95.08 %       0/0     0.00      NaN   2562       0/0      0.00   16.55 % 
 
 
 




Edited, May 8th 2011 11:30am by lynnminmay
#95 May 08 2011 at 6:12 AM Rating: Decent
**
983 posts
So, /WAR and Saber Dance . . . . I was really hoping to prove myself wrong here, but no luck.
Plot, T=0 is Saber Dance usage.
5 minute usage, +0% from gear, /WAR

1609/7679 (20.95% +/- 0.91%)

So DA from trait absolutely does not stack with Saber Dance, but DA from gear absolutely does. Balls!
____________________________
Yay for Jhereg!
http://www.ffxiah.com/player/Lakshmi/Byrth
#96 May 13 2011 at 10:11 AM Rating: Decent
*
52 posts
I'm interested on previous parsed results posted Page1 about the hypothesises:
- DA check before TA
- TA check before DA
... with "fixed" KParser.

Could testers re-analyze their parses and post new results here please ?
#97 May 13 2011 at 11:10 AM Rating: Decent
**
415 posts
It's 4 posts aboves yours.
#98 May 13 2011 at 12:14 PM Rating: Decent
**
983 posts
I explained it in the BG thread where I linked to it. It was definitely a very solid testing idea.

What do you guys think of this for a final Saber Dance test:
1) Max Haste and DW gear (48% DW, ~26% Haste)
2) 2x Dagger (200 delay)
3) Activate JA once every 6 minutes and do literally nothing between

Then I can extract the timestamps for attacks and Saber Dance usage from the logs. I subtract the most recent Saber Dance usage from the attack round timestamps, add up the attacks that happen at the same time latency from Saber Dance, and plot it.

Minute 5-6 is the baseline Attack rate. I average the number of attacks at those time points and divide the entire plot by that number. It should reduce it to the percentage increase in my number of attack rounds by Saber Dance. If I start with 0% DA, this should be my Double Attack rate. It's going to be a little more fuzzy than that, because DA rounds appear in the chat log with some latency, so my resolution can't really exceed that latency, but this would be a much faster way to collect data. I can have my mule there on Bard using Marches and Haste.
____________________________
Yay for Jhereg!
http://www.ffxiah.com/player/Lakshmi/Byrth
#99 May 13 2011 at 1:42 PM Rating: Good
*
52 posts
ok Pchan, sorry i missed it (not used to the expand feature on this forum < <)

Good Job and i <3 the name yu chose for yur mule XD

EDIT: what were the total fighting time exactly (for your test with Signet Staff delay366)? Same question for your LadyBell test (delay203 Haste6%).
I'm trying to retrieve the number of rounds and doing so the correct divisor (60 or 59 or 58.85 or something else ?)
Also, i think (but not sure) found another bug with KParser might display wrong total times.

EDIT2: there is something else i don't understand about Impetus critrate enhancing feature: you saying +1% per crit, ok but description says resets if miss. So if i understand correctly, if you don't crit next hit = lose all your critrate bonus given from Impetus ? Unless im missing something, i don't understand there XD.

Edited, May 14th 2011 4:07am by Masamunai
#100 May 14 2011 at 5:04 PM Rating: Good
Scholar
46 posts
Quote:
EDIT2: there is something else i don't understand about Impetus critrate enhancing feature: you saying +1% per crit, ok but description says resets if miss. So if i understand correctly, if you don't crit next hit = lose all your critrate bonus given from Impetus ? Unless im missing something, i don't understand there XD.


I'm pretty sure he meant to say +1% per hit, whether it crits or not. It works just like the attack bonus which resets after a miss and rises with each hit.
#101 May 15 2011 at 5:12 AM Rating: Decent
**
415 posts
Masamunai wrote:
ok Pchan, sorry i missed it (not used to the expand feature on this forum < <)

Good Job and i <3 the name yu chose for yur mule XD

EDIT: what were the total fighting time exactly (for your test with Signet Staff delay366)? Same question for your LadyBell test (delay203 Haste6%).
I'm trying to retrieve the number of rounds and doing so the correct divisor (60 or 59 or 58.85 or something else ?)
Also, i think (but not sure) found another bug with KParser might display wrong total times.


https://rapidshare.com/files/2514721986/DA15TA10-mule.sdf

It only lists 5 min as a duration though... It lasted like 15 hours


Quote:

EDIT2: there is something else i don't understand about Impetus critrate enhancing feature: you saying +1% per crit, ok but description says resets if miss. So if i understand correctly, if you don't crit next hit = lose all your critrate bonus given from Impetus ? Unless im missing something, i don't understand there XD.

Edited, May 14th 2011 4:07am by Masamunai


yes you lose everything, it's a pretty markov chain.
Reply To Thread

Colors Smileys Quote OriginalQuote Checked Help

 

Recent Visitors: 19 All times are in CDT
Anonymous Guests (19)