Relics Xtimes damage proc rate testings

[ringthree]

It seems like you don't understand that crits and X procs are independant of each other. As sample size rises, it becomes irrelevant that you're just looking at crits instead of your whole set of attacks. You're trading off slower results in real time for the convenience of being able to do it in a merit party, personally I think it's a very smart way of doing it.

No, I don't think it is as simple as that. The problem is that the procs are independent of each other. Crits are an unknown sample size. We have only guess work to figure out the crit percentage, and different mobs have different stats causing different crit percentage. This means that the results will be pretty inaccurate compared to one person just sitting down and killing some rabbits.

Like I said, I understand why people are doing it this way, but in the end it is just as good as guesswork. Convenience is great, but convenience doesn't give you good statistics. Just because you people to participate because of the convenience isn't going to make the results anymore correct.

[Thorny]

No, I don't think it is as simple as that. The problem is that the procs are independent of each other. Crits are an unknown sample size. We have only guess work to figure out the crit percentage, and different mobs have different stats causing different crit percentage. This means that the results will be pretty inaccurate compared to one person just sitting down and killing some rabbits.

Like I said, I understand why people are doing it this way, but in the end it is just as good as guesswork. Convenience is great, but convenience doesn't give you good statistics. Just because you people to participate because of the convenience isn't going to make the results anymore correct.

Go take an introductory statistics course, thanks. I can't recall the last intelligent post you've made anywhere..

[ringthree]

I think what RT isn't understanding is that for 2h weapons the range is very skewed, and even 1h weapons sometimes (I remember having to pause and figure out if xxx hit was a 3x or not with bluefagdag)

Crits make it easier it's just a bit more tedious I guess? @25% base crit and 8% xhit rate you're looking at 1/50 hits where (for the purposes of mathffff) where you're going to get crits instead of 2/25 for the xhit damage.

What (i believe) mas is suggesting is to test on lower level shit where we're 100% sure of cap on crit to maximize the frequency. To be honest if you don't already have a stockpile of parses it's an annoying task.

To contribute to the thread, rounded off, 7.7% overall 3x proc rate on Colibris (kikoku, no info on brovura yet) (taken from 3 parses, not big sample size) and I have no contributing information for 3x crit frequency.

I just woke up so if I got the purpose of the thread wrong just slap me.

Actually, I am suggesting that you test on the lowest level stuff because the results will be obvious and actually will be easier to pick out of all the results and not just crits. The test as it is will get more data because it just requires a relic holder to merit, but even with more data the results will be guesswork.

[ringthree]

Go take an introductory statistics course, thanks. I can't recall the last intelligent post you've made anywhere..

You will be able to get the information that you want with a greater margin of error if you do the crit parsing. If you kill level 0 mobs the margin of error goes way down.

I didn't say it wouldn't work, I said it wouldn't give enough accurate data to give you a better idea than the work that has already been done.

[Thorny]

The margin of error is the exact same if you parse 100 crits versus 100 landed hits.. they are fucking independant of X-damage procs and random functions don't have memory so the fact you have landed hits in between changes NOTHING. The primary difference is that the critical method is going to be more productive for the person testing, albeit marginally slower to hit the same sample size depending on amount of haste buffs/critrate/etc. If this is too complicated for you to understand, soloing rabbits won't get you any haste buffs and you have to run between them which causes delay, while meritting you should always have a target and significant haste buffs, at 20% crit rate you're going to be getting pretty close to as many crits landed over time as you would get melee hits doing rabbits outside town.

Here, idiot's example for you:

You alternate flipping a nickel and a quarter. You only record the heads/tails on the quarter. Does your rate of heads change versus recording both coins(assuming they're fair coins and 50-50 heads/tails to begin with)?

[Kirschy]

I don't understand what you mean by the margin of error being different on Lv0 mobs vs merit party mobs.

The amount of relic procs in the group of melee hits should be equal to the amount of relic procs in the group of critical hits. The distribution should be equal between both groups. (If not, then of course this method is faulty.)

Assuming the distribution is equal, then just knowing the number of critical hits, and the number of procs in that pool should be sufficient. (Which is easy to record using KParser.)

The crit rate compared to the regular hits shouldn't affect anything. An analogous situation would be parsing hits for 2hours, but only choosing the sample size of the last 30 minutes of parsing. The distribution in the first 1.5 hrs = distribution in the last .5 hours. This works for any time x. Assume a parse for y hours, the distribution of hits after the first x minutes of parsing is equal to the distribution of the remaining y-x minutes. The ratio of y plays no role in determining the rate of proc.

[ringthree]

The margin of error is the exact same if you parse 100 crits versus 100 landed hits.. they are fucking independant of X-damage procs and random functions don't have memory so the fact you have landed hits in between changes NOTHING. The primary difference is that the critical method is going to be more productive for the person testing, albeit marginally slower to hit the same sample size depending on amount of haste buffs/critrate/etc. If this is too complicated for you to understand, soloing rabbits won't get you any haste buffs and you have to run between them which causes delay, while meritting you should always have a target and significant haste buffs, at 20% crit rate you're going to be getting pretty close to as many crits landed over time as you would get melee hits doing rabbits outside town.

Here, idiot's example for you:

You alternate flipping a nickel and a quarter. You only record the heads/tails on the quarter. Does your rate of heads change versus recording both coins(assuming they're fair coins and 50-50 heads/tails to begin with)?

I get what you are saying, but your example is wrong. If you rolled a 100 sided-die 100 times looking for an event that happens 5% of the time but only took the results of 20% of the rolls as results you have to get a huge number of results to equal the margin of error on just looking for the 5% in 100 rolls. Basically, all I am saying is that it's easier to find the percentage from 100% of the results than less than 20% of the results, this should be obvious.

You are really getting riled up over nothing. Of course it's going to be more productive to do this while meriting. I have no problem with that. The results just aren't going to be as good. I guess we just disagree on the method.

[ringthree]

I don't understand what you mean by the margin of error being different on Lv0 mobs vs merit party mobs.

The amount of relic procs in the group of melee hits should be equal to the amount of relic procs in the group of critical hits. The distribution should be equal between both groups. (If not, then of course this method is faulty.)

Assuming the distribution is equal, then just knowing the number of critical hits, and the number of procs in that pool should be sufficient. (Which is easy to record using KParser.)

The crit rate compared to the regular hits shouldn't affect anything. An analogous situation would be parsing hits for 2hours, but only choosing the sample size of the last 30 minutes of parsing. The distribution in the first 1.5 hrs = distribution in the last .5 hours. This works for any time x. Assume a parse for y hours, the distribution of hits after the first x minutes of parsing is equal to the distribution of the remaining y-x minutes. The ratio of y plays no role in determining the rate of proc.

The distributions should equal in a smaller sample compared to a larger one. If you think it's faster to merit, then so be it.

[Thorny]

I get what you are saying, but your example is wrong. If you rolled a 100 sided-die 100 times looking for an event that happens 5% of the time but only took the results of 20% of the rolls as results you have to get a huge number of results to equal the margin of error on just looking for the 5% in 100 rolls. Basically, all I am saying is that it's easier to find the percentage from 100% of the results than less than 20% of the results, this should be obvious.

You are really getting riled up over nothing. Of course it's going to be more productive to do this while meriting. I have no problem with that. The results just aren't going to be as good. I guess we just disagree on the method.

The results ARE going to be as good. If you're trying to say you need a higher sample size overall, you're correct. However, parsing 100 critical hits in merit will have the same effect as parsing 100 plain swings in ronfaure(if you don't agree with this, I give up on you).. the difference in time spent isn't as big as you would think it is, given that in merit you're going from 35% haste to 70/80% depending on dnc or not and have less disengaged time. As long as you understand that your sample size is your #/critical hits and not your total number of swings in the merit example, they're the exact same. The upside is that doing it in merits is more productive outside of the test and requires less effort to retrieve your data afterwards.

[Masamune]

You will be able to get the information that you want with a greater margin of error if you do the crit parsing. If you kill level 0 mobs the margin of error goes way down.

I didn't say it wouldn't work, I said it wouldn't give enough accurate data to give you a better idea than the work that has already been done.

RT: suppose you are a potential contributor owning a relic, you read this testing 1st paragraph :

- slap 500 lvl0 mobs, running around to kill them one by one mindlessly, and get full data with corresponding margin of error.

or

- just get out your parses already done off yur storage dust, and yur handcalc + a paper/pen + 30 mins of your time to post here. It appears you have a few lolibris parses just for proving yur epeen/improving yur gears to use yur relic when yu got it some time ago, with barely 50 procs each.

Which option will you run away and other maybe do the effort to contribute ?

the whole accuracy arguement is only the analyzers problem, not the contributor's, he don't fucking care besides the conclusion. The contributor fucking care also if such test will ask him to pass 2 weeks almost non stop testing, compared to 30 mins of his time.

The rest, it's us analyzers who will have ot make up/compensate with hoped more contributors than "slapping lvl0mobs" option which would prolly returns 0 and you would see this topic in same state as my RagingRush crit test @ KI and Allak: full of trolls or dead.

[Kirschy]

The distributions should equal in a smaller sample compared to a larger one. If you think it's faster to merit, then so be it.

Yea there's a chance they're not equal. Considering we've tested plenty of other aspects of the game and we're still +/-5% on relic proc rate after ~5 years shows there reluctance to spend hours to perform this kind of testing. (Even knowing the answer isn't going to change the way we play, we'll just have more accuracy when doing item comparisions =p.)

As one of the few people who have spent countless hours testing random aspects of the game from recast to accuracy, I can say that getting a ~10k sample size on Lv0 mobs is a lot harder than it seems. As it was mentioned, you spend more time engaging/running to the next mob than you do hitting. Even doing it on crabs is rough. (Just testing Mythic proc rate over ~500+ hits took a long time, getting near 10k would just suck .)

The crits in merit style situations is just an easy/fast way to get a fairly accurate number, and deduce if the relic proc rate is the same across all weapons~ I know you're not adamantly against any of this, I just feel I should defend the testing in hopes that people will contribute!

[ringthree]

The results ARE going to be as good. If you're trying to say you need a higher sample size overall, you're correct. However, parsing 100 critical hits in merit will have the same effect as parsing 100 plain swings in ronfaure(if you don't agree with this, I give up on you).. the difference in time spent isn't as big as you would think it is, given that in merit you're going from 35% haste to 70/80% depending on dnc or not and have less disengaged time. As long as you understand that your sample size is your #/critical hits and not your total number of swings in the merit example, they're the exact same. The upside is that doing it in merits is more productive outside of the test and requires less effort to retrieve your data afterwards.

Yes, my presumption is that the same size would be smaller. I am pretty sure that you need 80% haste constantly to be able to get the same amount of hits killing bunnies. It's definitely more boring though.

[ringthree]

RT: suppose you are a potential contributor owning a relic, you read this testing 1st paragraph :

- slap 500 lvl0 mobs, running around to kill them one by one mindlessly, and get full data with corresponding margin of error.

or

- just get out your parses already done off yur storage dust, and yur handcalc + a paper/pen + 30 mins of your time to post here. It appears you have a few lolibris parses just for proving yur epeen/improving yur gears to use yur relic when yu got it some time ago, with barely 50 procs each.

Which option will you run away and other maybe do the effort to contribute ?

the whole accuracy arguement is only the analyzers problem, not the contributor's, he don't fucking care besides the conclusion. The contributor fucking care also if such test will ask him to pass 2 weeks almost non stop testing, compared to 30 mins of his time.

The rest, it's us analyzers who will have ot make up/compensate with hoped more contributors than "slapping lvl0mobs" option which would prolly returns 0 and you would see this topic in same state as my RagingRush crit test @ KI and Allak: full of trolls or dead.

If I had a relic, the best you were going to get out of me is Einherjar parses. I only have time for events, and Omoi makes me go WHM to merit parties in the rare event that we actually do one.

[Thorny]

Yes, my presumption is that the same size would be smaller. I am pretty sure that you need 80% haste constantly to be able to get the same amount of hits killing bunnies. It's definitely more boring though.

With capped gear haste and hasso, you're going from 35% haste to 70% comparing normal to merit, 2.16x more attacks in merit. If you spend twice the engaged time(pretty likely with all the running around, but that's more of an approximation), you'd need a 23% critrate to get the same amount of samples over time(so you'll get slightly fewer over time meritting but far more likely to do it for extended periods of time). If you assume dnc, you're getting 3.25x more attacks and only need a 15.3% critrate assuming the same twice engaged time.

[Masamune]

Less chit chat, let the relic owners talk please.

To contribute to the thread, rounded off, 7.7% overall 3x proc rate on Colibris (kikoku, no info on brovura yet) (taken from 3 parses, not big sample size) and I have no contributing information for 3x crit frequency.

I just woke up so if I got the purpose of the thread wrong just slap me.

You partially got it actually XD, we missing the #tripleprocs/#crits = 7.7%.

/em superslapshake you and gives you coffee. :D

[Vodou]

Not to be all agreeable, but Mas, Thorny and RT are all making good points on the method of collecting this.

1) yes more sample size will be critical to getting good results (RT)

2) yes, criticals are just as good as normals (assuming adequate sample size) (Thorny)

3) yes, pre-existing parses and no downtime massacring bunnies is much more likely to get results (Mas)

Just wanted to say so.

Sadly I don't have a lot on merits, and the data I have is flawed (Dark Ixion) since a side shot triple crit looks the same as a rear-shot normal criitical. (If you think you can somehow use that data, I can post.)

I might do bunnies tho, since I love seeing a triple proc for 1700.

I'll see what I can do to contribute.

I assume we can accumulate results from multiple parses from different people (provided its the same relic and can determine the data to be useable.) I'll get what I have posted when I get home.

[Kaeko]

Assuming that indeed the "Relic Proc Rate" (what you are attempting to find) is unaffected by the "Critical Hit Rate" - that is increasing or decreasing the critical hit rate does not change the basic relic proc rate, then Masamune's proposed methodology should work.

It should not matter what the Critical Hit Rate is in this situation due to the fact that the 2 events are independent. Using 2 extremes to illustrate this, your relic proc rate should be the same if your critical hit rate was 1% or 90%. Your raw data will look different; however, as long as you are able to clearly discern between the 4 possible outcomes of a single hit:

(1) normal hit
(2) relic hit
(3) critical hit
(4) critical + relic hit

then you will be fine. A method of this was already proposed and generally accepted (that the possibility of an "overlap" between relic hits and critical hits is non-existent due to PDIF caps and minimums not overlapped). Changing the critical hit rate will only increase the frequency of (3)+(4); however, it will not increase ratio of (3) to (4), which is what you want.

[Masamune]

just got an idea: i know an apoc and a mandau users, if i could get meripos with them and parse them that would work ?

[Kaeko]

Going along with the theme of testing methodology, you have to always ask yourself is your data collection actually feasible. That is, can you actually get enough adequate data using this method? Because you are dealing with a relatively low "success" on relic proc, you are going to require a pretty large number of attempts.

You can easily get thousands of swings with parses with little to no effort. You have to spend hours, even days, collecting this on bunnies. One is just not feasible - you will be hard pressed to find people with relics that are willing to do this to that extent. Hitting bunnies does indeed control for more possible confounding variables, but it's not feasible so it's not a good test. Parse data controls for less potential confounding variables, but there is some logic behind not controlling for said variables.

Sometimes you have to sacrifice precision for practicality.

[BerenTebogo]

Less chit chat, let the relic owners talk please.


You partially got it actually XD, we missing the #tripleprocs/#crits = 7.7%.

/em superslapshake you and gives you coffee. :D

I don't have access to the raw data anymore for a long time now but I remember, as I said rounded off, the triple proc rate was about 7.7%

I know plenty of people out there can do actual tests to give the exact number but it's the best I can do with only a stage 4 on me and my full relic long gone ;;