After thinking about it... really stupid question.
A pity that its such a big logfile with group fluctuations / party buff changes. Else we could do some calculations on glancing dmg loss from it.
But i guess the paladin party could solve that issue for us.
After thinking about it... really stupid question.
A pity that its such a big logfile with group fluctuations / party buff changes. Else we could do some calculations on glancing dmg loss from it.
But i guess the paladin party could solve that issue for us.
Glancing damage penalty is easy to find, you found the hard part, the % chance to glance.
Now equip a low level vendor weapon with a small damage range (2 or 3) and get a glance or two on a boss to find your penalty.
Not sure why I have more swings than you, but 18MB was too large for my poor parser, so I first split up the combatlog using a modified version of your parser.bat. For Kael'Thas I had to use only Kael though, because else it yielded 0 results.
Could be a bit error prone here.
Anyway, from this chart it looks like the glancing blows did 76.76% damage of a normal white swing, so a rough 24% reduction in damage.
Crit Difference is .793%, corrected to .673% to account for agi gap, with a std dev of .188%. Hence, with a high degree of confidence, the true number lies within 1.041 and .305. Assuming the skill difference is 15 as previously asserted, the true crit per skill number lies between .02% and .06%.
Same calculation for hit: hit per skill lies between .106% and .181%.
Same calculation for dodge: dodge reduction per skill lies between .053% and .112%.
Conclusion? Well, the first one is that the data set would need to be about 5 times larger to really pin this down with much accuracy. But .04% crit per skill looks pretty good, .1% hit is now on the outliers, and dodge has the look of some oddity like .08%
Aldriana for us math inepts.. can you rephrase Crit Difference, or explain what you are saying a little bit more in depth. I have reread that post a few times and I am still having a hard time deciphering what exactly you are saying about crit ;-P
Aldriana for us math inepts.. can you rephrase Crit Difference, or explain what you are saying a little bit more in depth. I have reread that post a few times and I am still having a hard time deciphering what exactly you are saying about crit ;-P
Thanks in advance.
So, the idea with statistics is that, with random occurances, you can have runs of good luck or back luck, such that the the percentage observed from a given data set will generally not agree precisely with the theoretical value. What statistics does is provide bounds on how far, on average, it will stray from the theoretical value; for instance, if you flip a coin 1000 times, on average, you'll get 500 heads; but, on a given set of 500 heads, you may get 480 or 510 heads, but you'll never every see 200 or 700 heads (assuming the coin is fair). Statistics lets us tell where the break between "close enough that the deviation is probably just random variation" and "too far away to be explained by random variation alone"
So in this case, I worked out the average and std. deviation (which is a statistical value related to how much the data is allowed to spread); they turned out to be .793% and .188%. Now, of the difference in crit value, .12% is explained by the difference in agility; this means that .793% - .12% = .673% of the difference is explained by weapon skill.
Now, statistics tells us that, in 95% of cases, observed values via a test will lie within 1.96 standard deviations of the mean. So, we compute .673% + 1.96*.188%, and .673% - 1.96*.188%, and find that there is a 95% chance that the true difference in crit value accounted for by weapon skill lies between .305% and 1.041%.
Then, since the weapon skill difference was 15, we divide these numbers by 15 to get the amount of crit granted by each point of weapon skill. Doing so indicates that the amount of crit given by +1 weapon skill is between .0203% and .0694% (the .02 to .06 I wrote last night was a typo; I meant .02 to .07).
Hence, the true amount of crit per weapon skill is 95% likely to be between .0203% and .0694%, based on this data set.
(And for those of you that know statistics, yes, I realize this is technically an abuse of statistics. But it's a useful estimating tool, and a whole lot easier than doing the totally correct statistical testing.)
Oops. I just went over my parser again and my "hits" are counting glancings and blocks for some reason too. I am going to fix it, post what I am using for a parser, and post the logs in just a second, but it screws up our results I posted earlier. Also, Theras can hopefully post the paper doll stats of the characters used.
So, the idea with statistics is that, with random occurances, you can have runs of good luck or back luck, such that the the percentage observed from a given data set will generally not agree precisely with the theoretical value. What statistics does is provide bounds on how far, on average, it will stray from the theoretical value; for instance, if you flip a coin 1000 times, on average, you'll get 500 heads; but, on a given set of 500 heads, you may get 480 or 510 heads, but you'll never every see 200 or 700 heads (assuming the coin is fair). Statistics lets us tell where the break between "close enough that the deviation is probably just random variation" and "too far away to be explained by random variation alone"
So in this case, I worked out the average and std. deviation (which is a statistical value related to how much the data is allowed to spread); they turned out to be .793% and .188%. Now, of the difference in crit value, .12% is explained by the difference in agility; this means that .793% - .12% = .673% of the difference is explained by weapon skill.
Now, statistics tells us that, in 95% of cases, observed values via a test will lie within 1.96 standard deviations of the mean. So, we compute .673% + 1.96*.188%, and .673% - 1.96*.188%, and find that there is a 95% chance that the true difference in crit value accounted for by weapon skill lies between .305% and 1.041%.
Then, since the weapon skill difference was 15, we divide these numbers by 15 to get the amount of crit granted by each point of weapon skill. Doing so indicates that the amount of crit given by +1 weapon skill is between .0203% and .0694% (the .02 to .06 I wrote last night was a typo; I meant .02 to .07).
Hence, the true amount of crit per weapon skill is 95% likely to be between .0203% and .0694%, based on this data set.
(And for those of you that know statistics, yes, I realize this is technically an abuse of statistics. But it's a useful estimating tool, and a whole lot easier than doing the totally correct statistical testing.)
Beautiful. I almost want to go back to college and brush up on math. Almost.
Alright, first, the new correctly parsed data. I should have posted the log in the first place, my mistake.
Therasiv:
327 crits, 4845 hits, 442 blocks, 2369 glances, 521 misses, 323 dodges, 1148 parries, 9975 total attacks
533359 damage, 53.469574 damage per swing
347863 hit damage, 71.798349 hit damage per swing
127203 glance damage, 53.694808 glance damage per swing
3.278% crits
48.571% hits
4.431% blocks
23.749% glancings
5.223% misses
3.238% dodges
11.509% parries
Beaglej:
196 crits, 3933 hits, 396 blocks, 2030 glances, 692 misses, 419 dodges, 1016 parries, 8682 total attacks
415263 damage, 47.830339 damage per swing
272925 hit damage, 69.393593 hit damage per swing
105224 glance damage, 51.834483 glance damage per swing
2.258% crits
45.301% hits
4.561% blocks
23.382% glancings
7.971% misses
4.826% dodges
11.702% parries
Here is the awk script used to generate this data. I added in the glancing damage for people that seemed interested in glance rates. Also, sorry for spreading the 18-19% glancings before as you can now see they are 24%. I am editing out my original post too so people don't read the mistaken data later and think it is correct.
Not sure how you determined the average and std. deviation however, if I assume that they are the same regardless and I am understanding you correctly we are looking at the following...
Um, close. Standard deviation is actually affected by the size of the data set; the larger the data set is, the smaller the standard deviation gets. I'll post the new calculations in a bit; I have some work I need to get done this morning before I waste any more time theorycrafting .
Um, close. Standard deviation is actually affected by the size of the data set; the larger the data set is, the smaller the standard deviation gets. I'll post the new calculations in a bit; I have some work I need to get done this morning before I waste any more time theorycrafting .
Sorry for the extra work. Checking my bandwidth usage on my server for this month. Will probably just post the log there in a bit if I am doing alright.
Um, close. Standard deviation is actually affected by the size of the data set; the larger the data set is, the smaller the standard deviation gets. I'll post the new calculations in a bit; I have some work I need to get done this morning before I waste any more time theorycrafting .
Can't fault a guy for trying... or can you.
Believe it or not, sadly I am sitting here waiting for a response as it's a slow work day for me. I need something to ponder.
Here, the log file is really small when zipped up. I only have data in it strictly from Venoxis (grep Venoxis), and I removed the small portion where Beaglej's weapon broke.
So, brief explanation of how to come up with standard deviations:
First, we come up with the standard deviations of each individual data set. The formula for this is:
sqrt(crits * (total - crits)/total)/total.
So, for instance, the standard deviation for Thera is sqrt(327 * (9975-327)/9975)/9975 = .178%, and the standard deviation for Beagle is sqrt(196*(8682-196)/8682)/8682 = .159%.
So, how do we compute the standard deviation of the difference? Well, as it turns out, when you add or subtract distributions, a statistical quantity known as the variance just adds directly. Fortunately for us, the variance is easy to calculate - it's just the standard deviation squared. So the standard deviation of the combined distribution is sqrt(.178%^2 + .159%^2) = .239%.
Thus, the difference between their crit values is 1.02%, with a variance of .239%. Running through the above calculations, we find that the crit per skill point is then 95% likely to lie between .0288% and .0912%.
Computing the same for miss rate, the range is .135% to .231% per weapon skill.
Computing the same for dodge rate, the range is .068% to .144%.
Conclusion: need more data. Although the base stats of the pallies would help narrow things down a bit.
So, brief explanation of how to come up with standard deviations:
First, we come up with the standard deviations of each individual data set. The formula for this is:
sqrt(crits * (total - crits)/total)/total.
So, for instance, the standard deviation for Thera is sqrt(327 * (9975-327)/9975)/9975 = .178%, and the standard deviation for Beagle is sqrt(196*(8682-196)/8682)/8682 = .159%.
So, how do we compute the standard deviation of the difference? Well, as it turns out, when you add or subtract distributions, a statistical quantity known as the variance just adds directly. Fortunately for us, the variance is easy to calculate - it's just the standard deviation squared. So the standard deviation of the combined distribution is sqrt(.178%^2 + .159%^2) = .239%.
Thus, the difference between their crit values is 1.02%, with a variance of .239%. Running through the above calculations, we find that the crit per skill point is then 95% likely to lie between .0288% and .0912%.
Computing the same for miss rate, the range is .135% to .231% per weapon skill.
Computing the same for dodge rate, the range is .068% to .144%.
Conclusion: need more data. Although the base stats of the pallies would help narrow things down a bit.
I am trying to get Theras to post those base stats.
Yeah, I would like some more data. Sadly, Venoxis was much stupider than we thought. He uses Holy Fire and Renew until he is out of mana. Then when he gains enough mana back, he uses it for Holy Fire instead of Renew. As such, once he is out of mana, he doesn't heal and we were able to push him to phase 2. This resulted in about 50 mins of testing then we had to wipe and restart it. Lose some time to killing the snakes too.
You wanna know who NEVER ran out of mana? Friggin' Yauj. ...
Friggin Yauj.
Yauj has an enrage timer though.
No, if we were going to get a group to clear to a boss, the absolute best on would be Ebonroc. Definitely can go forever against him. Well, at least until weapons break. Maybe bring a backpack full of repair bots to use.