Ranger ac parsing (I'm serious)

[DeathsSilkyMist]

Originally Posted by Duik [You must be logged in to view images. Log in or Register.]

So (to me it's) trust the data given (even if it is small in some peoples opinions) until a more robust set is provided.
But how big is big enough?

I would say samples have merit when multiple samples from different mobs match a proposed math formula. P99 uses math formulas for all of the combat calculations, including the chance of being hit. Once you know the math formula, you can apply it to every scenario the math formula is designed for, unless there is some exception in the code somewhere. While that is always possible, it is typically bad practice to add these exceptions coding-wise. Sticking to the existing math forumlas as much as possible is more consistent and easier to debug.

I think my post about the player melee damage formula is a good example of samples matching a proposed math formula:

https://www.project1999.com/forums/s...&postcount=440

I proposed a math formula based on EQEMU code, and got parse data from a few different mobs of varying level (and different expansions). The data matched the math formula fairly closely, and the sample size of each mob tested wasn't huge. The Epic Fist + SoS data was a sample size of roughly 1000 hits/misses per mob if I recall correctly.

It isn't as likely multiple samples from different mobs in different expansions would just so happen to line up with my formula by chance. Obviously this isn't a guarantee I am correct, and people can try to disprove my post. But I think it's a good starting point for determing if a sample size has merit, without assuming you need a large sample size like 1 million hits/misses.

[Snaggles]

Originally Posted by Vear99 [You must be logged in to view images. Log in or Register.]

Who are you talking about? I was interested enough in the OP to run my own experiment, and if you want to tank 10,000 hits on a L55+ NPC with multiple AC levels and classes and do all of the appropriate confidence interval math, I'd be very interested.

He was responding to this:

Originally Posted by Snaggles [You must be logged in to view images. Log in or Register.]

This is a sample of a lot of hits but there still is a chance of the relatively small sample pool providing scatter. Someone like bcbrown could probably use the right terms for what I’m trying to say.

Just like the law of large numbers tends to stabilize statistics, sometimes shorter runs can lead to sporadic results. That doesn’t villainize an attempt but it can mean the test itself is less illustrative than one would prefer.

I don’t intend to knock anyone’s effort to understand this game. Literally all my own testing is me trying to read the tea leaves because my samples are far too small. The closest thing to running a 10k parse like you said is just several Yeli parses averaged.

[Snaggles]

Originally Posted by Snaggles [You must be logged in to view images. Log in or Register.]

He was responding to this:

Actually he wasn’t now that I re-read that lol. Oops, sorry.

I’m sure the person “in the past” was not this thread. Most of us spin in circles and get a bit heated so it could have been anything [You must be logged in to view images. Log in or Register.]

[bcbrown]

Originally Posted by Vear99 [You must be logged in to view images. Log in or Register.]

Who are you talking about? I was interested enough in the OP to run my own experiment, and if you want to tank 10,000 hits on a L55+ NPC with multiple AC levels and classes and do all of the appropriate confidence interval math, I'd be very interested.

My apologies. I can see how you might have thought I was talking about the data you provided, but that was not my intention. I haven't really had enough of a chance to look at any of the data in this thread to form an opinion on how conclusive it is. That's part of why my response to Snaggles was purely theoretical (coin flips) and not practical.

When you provided your data you included the appropriate caveats about sample sizes. If you were willing to provide me with the raw logs of hits and misses I'd be very interested in taking a look.

I wish I could do that experiment myself, but my main is a druid with about 141 worn ac, and I don't think anyone is all that interested in how well a druid can tank. If I could find a good constant-level subject for my 30 bard I could run some tests from 100-200 worn ac. When I get my ranger to 50 (47 now) I plan to run some tests against a froglok hunter in Trakanon's Teeth, but he only has about 140 worn ac so I'm not sure how useful that will end up being. I've got a 54 cleric with close to 200 worn ac I'd like to parse against a froglok hunter as well.

If anyone ever wants to run a long parse, I'm happy to provide heals on my cleric on either blue or green, but no one's ever taken me up on that.

[Duik]

So (to me it's) trust the data given (even if it is small in some peoples opinions) until a more robust set is provided.
But how big is big enough?

This scenario is really only applicable to a p99 server locked in one finite min/max era.

There is a literal max HP available.
There is a max mana and AC etc.

Many on live would not have had the option to test 3 BIS items for a given slot to even be able to discern a difference large enough to make an impact. (Given that they most likely went on to luclin etc).

Seems weird to me that a 55AC bp with 100hp (in this era) has much less/more impact than a 45AC bp with AoB.
Either would have been fine for all but the MT back in 2000. Guilds went to the Nth degree to outfit an MT for good reason.
I do enjoy the probability discussions though.

[Snaggles]

My play time is a bit iffy these days but I’d offer to help heal as well. I wonder what would be a good target to plunk on. Maybe like a Freeport gate guard?

Note: https://wiki.project1999.com/Guard_Jacsen

Similar minimum to max hit ratio (2.7x) and level as some ToV npcs but far more sustainable for testing.

[Vivitron]

It looks like 769 would be absolute max worn ac on a bard; 674 without held items. That paladin's 538 is probably within the ballpark of what I would get with bis-ish gear viewing ac as worthwhile but tertiary to hp and resists.

If I count how many times I got hit for each specific amount of damage this is what I get:

     37 24
      5 30
      2 37
      3 44
      4 51
      9 57
      4 64
      4 71
      1 78
      2 85
      2 91
      3 98
      3 105
      6 112
      4 119
      5 125
      3 132
      1 139
      4 146

It would be interesting to compare this distribution to a parse against a lowbie mob (just high level enough to have 20 distinct hit buckets*), and also against parses vs desired targets.

It would also be interesting to read the eqemu source and see if the behavior matches; if so we may have an exact answer there.

* I note I only got hit for 19 different values here.

[bcbrown]

Originally Posted by Vivitron [You must be logged in to view images. Log in or Register.]

If I count how many times I got hit for each specific amount of damage this is what I get:

It would be interesting to compare this distribution to a parse against a lowbie mob (just high level enough to have 20 distinct hit buckets*), and also against parses vs desired targets.

This is the analysis that I think will prove the most fruitful. Looking at average hit can provide hints, but looking at the full distribution has a lot more detail.

My understanding of the damage calculations is that it's meant to be a normal distribution with some mean that gets shifted around the DI values depending on the ac/atk comparison, with tails clipped. The number of hits you recorded obviously isn't enough for definitive conclusions, but just eyeballing it I bet 57 is the mean value. Since it's so far below the middle value of the DI, the clipping on the lower tail provides a huge number of minimum hits. I bet if you took a longer parse, the missing 20th value would be the highest number, 152 or 153.

So your AC is fairly completely blanketing the mob's attack value here. Obviously it would be way too much work for anyone to do, but a really interesting experiment would be to play with the worn ac value until you get the mean hit to be the 10th and 11th values of the DI of 85 and 91, then finding how much worn ac you need to add/remove to move that up or down by one AC interval.

[sogundordor]

I keep reading this post but quite busy working in these days to reply~
about the distribution thing,
the following chart shows the hits distribution of 97 worn ac:
[You must be logged in to view images. Log in or Register.]

this chart is shows 1396worn ac:
[You must be logged in to view images. Log in or Register.]

*Higher ac = higher chance for minimum hit = lower chance for max hit
*Lower ac = lower chance for minimum hit = higher chance for max hit
*no idea how those middle value works

Maybe how ac work is very simple~ just control the chance of min/max hitting =P

*I excluded bash/kick lines from all of my analysis too~

[Goregasmic]

Originally Posted by sogundordor [You must be logged in to view images. Log in or Register.]

Maybe how ac work is very simple~ just control the chance of min/max hitting =P

That was understood long ago I think. The higher the AC the lower you'll sit on the distribution. It seems like defense plays a big role in avoidance though and rangers lower def compared to knights and warriors is possibly where it hurts a lot, considering the AC soft cap is supposed to be the same for everyone.