Taking the median for WS damage?

Status
Not open for further replies.
C

chupunpupun

Guest
Non-Validated User
Is it possible to record each WS damage then calculating the median instead of taking the average?

I'm asking this because my Rampage marco have a two second delay for equip change and sometimes by the time rampage is used, the mobs HP is at 2% and my rampage would be like 40-60. DVS parse would record that as my low and skew my average.
 
Well Median is just (Max+Min)/2 right? If the average was skewed because of a low Min, the Median will be skewed as well.

However, I didn't think WSs worked this way. If the mob has 200HP left and you do a WS for 500, it will still say 500 in the chat log, not 200. I could be wrong about this, but this is how I thought it worked. That's why you can go Blade: Jin a Wild Rabbit in Ronfaure and do over 100 damage, even though it didn't have 100HP to begin with.
 
Kalia said:
Well Median is just (Max+Min)/2 right? If the average was skewed because of a low Min, the Median will be skewed as well.

However, I didn't think WSs worked this way. If the mob has 200HP left and you do a WS for 500, it will still say 500 in the chat log, not 200. I could be wrong about this, but this is how I thought it worked. That's why you can go Blade: Jin a Wild Rabbit in Ronfaure and do over 100 damage, even though it didn't have 100HP to begin with.
Click to expand...

It has to do with multi-hits. You can do more damage than the mob has life, but if you kill it with your first hit, *none of the other hits lands*, causing very low damage to appear. Rampage is a big one for this, for BST especially, because we want to release at very low life. You might do a 200 damage Jin on a ronfaure rabbit but you'll never do a 900 damage one like you'll see on XP mobs.

BTW math nitpick, median isn't related to min or max directly. It's just the middle number when you order the set of data. Or the average of the two if there are two "middle" numbers. In the situation specified, median would still be skewed. But maybe by less.... there's really no good way to model your average 5-hit rampage damage unless you only WS when there's a lot of life left on the mob.

-benny

edit, this is just randomly out there, but conversely this means that 1-hit WS will always show consistent damage on any mob at any life total, which is why people used them a lot to test the TP modifier of the WS gorgets.
 
It goes based off of every hit that lands. For instance, if you toss a Dancing Edge onto a lvl 1 bunny as a 75THF, the first hit will be recorded as doing full damage. However, every other hit in the 5-hit WS will "miss", because the mob is dead after the first hit, giving you 5% TP return and only somewhere around 500dmg.

*EDIT* I are to slow.

Oh and as for the median thing... it actually would work well to kinda of disregard fluke low numbers from things like Flash or a multi-hit at low mob health. Think of it this way:

You WS 5 times, with most of your skills landing for between 800 and 1000. However, one time you get hit with Flash and land a WS for 100dmg.

800, 900, 900, 1000, 100

The mean for this set would be 740, while the median would be 900. Pretty big difference if your typical WS values follow a standard distribution.
 
Last edited:
The problem Nek is that there's going to be a lot of numbers in between too.... 2-hit, 3-hit, 4-hit rampages. Certainly there are 5 distributions being added here and certainly with enough work you could more or less separate them, but not really within the scope of a mainstream-style parser.

It's all for WAR e-peen anyway (._.)

-benny
 
benny said:
BTW math nitpick, median isn't related to min or max directly. It's just the middle number when you order the set of data. Or the average of the two if there are two "middle" numbers. In the situation specified, median would still be skewed. But maybe by less.... there's really no good way to model your average 5-hit rampage damage unless you only WS when there's a lot of life left on the mob.
Click to expand...

Oops, major doucheness on my part. Guess that's what happens when you try to think about math while meriting lol. Another option would probably be to include an option to ignore Weaponskills that kill the mob. That would actually be very possible and not terribly hard to implement.
 
Benny said:
The problem Nek is that there's going to be a lot of numbers in between too.... 2-hit, 3-hit, 4-hit rampages. Certainly there are 5 distributions being added here and certainly with enough work you could more or less separate them, but not really within the scope of a mainstream-style parser.

It's all for WAR e-peen anyway (._.)
Click to expand...

I see what you're saying. I wasn't actually thinking of implementing a function to separate those WS out, I was just commenting on the fact that reporting the median WS would be a valuable statistic when fighting things with Flash, Stoneskin, etc. The problem is, a mean can be calculated off the running damage total (which is what I assume Kalia has done) while the median can only be reported by going back and creating a list of every WS performed by the player and sorting by damage done after every weaponskill performed. I doubt that would be feasible for a real-time parser like Kalia's.
 
Nekio said:
the median can only be reported by going back and creating a list of every WS performed by the player and sorting by damage done after every weaponskill performed. I doubt that would be feasible for a real-time parser like Kalia's.
Click to expand...

It wouldn't be THAT bad actually. The version that is out currently keeps all statistics in memory while the program is running, so each new thing I want to keep track of adds a level of complexity and resource usage. The next version will move to a database system where a small number of basic statistics will be stored in a database and queried/manipulated when it's time to display something to the user. The advantage of this is the flexibility you get with a database. It's very easy for me to store every single hit anyone has ever performed in chronological order, query them based on any number of conditions, and find the number in the middle for example. I'll have to do some testing and tweaking to see if it's going to be feasible due to the size of the tables that would produce, but it's definitely possible, and not that hard.
 
I also just thought of something else for down the line. Every time a WS is used, fetch the player's TP from memory (just like TParty does), and provide an advanced view that graphs Average WS damage as a function of TP return. That would be awesome.
 
I think excluding a multi-hit WS that results in a kill shot is silly. You'd end up excluding 1k+ dmg WS when the mob was 30% HP, etc. If someone wants to waste TP/lower his avg WS by rampaging a mob that's at 2% health that's on them. They should just take the next melee hit that'd kill it anyway and ws on the next mob. Let players play smarter rather than manipulate the parse to make them look better.
 
Fodder said:
I think excluding a multi-hit WS that results in a kill shot is silly. You'd end up excluding 1k+ dmg WS when the mob was 30% HP, etc. If someone wants to waste TP/lower his avg WS by rampaging a mob that's at 2% health that's on them. They should just take the next melee hit that'd kill it anyway and ws on the next mob. Let players play smarter rather than manipulate the parse to make them look better.
Click to expand...
QFT
 
Damn, I was gonna reply earlier but my DSL died at midnight and didn't come back on til I woke up....

I was gonna say that the median would give a more accurate view because it ignores the two extremes but mutiple people have already illstrated this. Anyway, I hope you don't remove the max because that data is relevent. The min is actually pointless because I can rampage for zero given the right coditions and that doesn't really tell me anything.
 
chupunpupun said:
Damn, I was gonna reply earlier but my DSL died at midnight and didn't come back on til I woke up....

I was gonna say that the median would give a more accurate view because it ignores the two extremes but mutiple people have already illstrated this. Anyway, I hope you don't remove the max because that data is relevent. The min is actually pointless because I can rampage for zero given the right coditions and that doesn't really tell me anything.
Click to expand...

Actually values of 0 are ignored by the parser (I think). There's not really any practical situation where a value of 0 does anything other than skew the results, so I went out of my way to have that case ignored since usually it just means you're hitting a mob that has Invincible or Physical Shield up. I'll have to double check exactly where and when 0s are ignored, but I know it's in there somewhere.
 
Kalia said:
Actually values of 0 are ignored by the parser (I think). There's not really any practical situation where a value of 0 does anything other than skew the results, so I went out of my way to have that case ignored since usually it just means you're hitting a mob that has Invincible or Physical Shield up. I'll have to double check exactly where and when 0s are ignored, but I know it's in there somewhere.
Click to expand...
That's cool as shit Kalia, no more Diamondhide screwing up the avg. :D
 
I know he's got it set to not count the ones that deadend into Invincible. I noticed that on the Dynamis-Bastok parse.

http://www.infieldpress.com/parser/1-Parser Logs/SavedParse--14_12_2006--0011.html

Nobody listed has a completely zeroed "dud WS" as the low even though I know I saw a few go into Invincible.

Diamondhide would still screw up your average by lowering it unless you never weaponskilled while it was up. I usually give up and dump it whenever I can. That stoneskin has to come off eventually.

http://www.infieldpress.com/parser/1-Parser Logs/08-13-2006.html

I know the lows listed on that parse were due to stoneskin. I don't believe that someone would WS for 6 or 7 DMG without ripping off the remaining stoneskin on their WS. That's what I'd expect to see though, that's the nature of the ability.
 
Last edited:
Yea, nobody listed has any dud non-WS melee attacks either, but it's a given that at least one mob used Invincible.

I should check how the parser handles that case. 0-damage attacks should probably still count towards Accuracy.
 
It's actually a very good idea to WS into Diamondhide, because melee hits for 0 return 0 TP, so you want to rip it off as fast as possible with strong damage.

I used to try to use elemental magic but melee hits drop it too fast for that to work, I found.

-benny
 
Kalia said:
Actually values of 0 are ignored by the parser (I think).
Click to expand...
Ya, WS going for 0 damage aren't being registered by the parser, which is nice when fighting WHM Mamools, or Hightrolls.
Benny said:
I used to try to use elemental magic but melee hits drop it too fast for that to work, I found.

-benny
Click to expand...
What you can do though, or usually RDM will, is throw a Sleep when they use Diamondhide, so the Hightroll will stay asleep until melees break it, giving tank a few seconds break.
 
If you really want a way to deal with those low WSs, a better solution would be to have two WS columns, a raw mean and an adjusted mean. The raw mean would be just that, the raw mean of all your WSs. This would include the 40 damage, single hit rampages, but also the "every hit lands and scores a critical and double attack procs" rampage. The adjusted mean column would adjust for outliers by removing them. Chauvenet's criterion is a pretty good one for outliers, but it's more work than it's worth if you ask me.
 
Status
Not open for further replies.
Back
Top