How to analyze loot

[Max Proteon]

Originally Posted by Snabble

heres how i analyze loots

Animal hide-- tt food - check
Animal oil -- tt food - check

etc etc

Originally Posted by falkao

hehe, its very precise

...and not to mention very simple

[falkao]

Originally Posted by Jimmy B

Yikes, missed this thread. Looks interesting falkao, I'll have a good read through later

Hi Jimmy, the thread is way to long now, so you might need some time. In the main post you'll find the basic stuff and some thoughts. The final model is depicted here model. However, I did not give any detailed explanation on how I did estimate it. First, to avoid further confusion, and secondly because I'm not sure if it's good to reveal all the secrets. Nevertheless, I'll provide further information via pm if requested.
Let me know what you think.

[Muppis]

I think you got it all kinda wrong in a sort of way, not to put you down or anything since i want to help

You have these nice models, which looks decent...but you want to find the peaks.

Gogo F´(x)? : )

And if i haven't forgotten my math, all those functions looks like they are exponential(english spelling?) functions or y = C * A^(x), atleast from the looks of http://www.entropiaforum.com/gallery...v_Loot1000.jpg

[falkao]

Originally Posted by Muppis

I think you got it all kinda wrong in a sort of way, not to put you down or anything since i want to help

You have these nice models, which looks decent...but you want to find the peaks.

Gogo F´(x)? : )

And if i haven't forgotten my math, all those functions looks like they are exponential(english spelling?) functions or y = C * A^(x), atleast from the looks of http://www.entropiaforum.com/gallery...v_Loot1000.jpg

The plot refers to normal loot, the mean of its gap (the gaps are not displayed) are exp distributed. However, I've included now the remaining part (globals, hof) and there are still the same distributions. If you read the first post, you'll see all ends up in mixtures of exp distributions.

Nevertheless, the above mentioned model does not model normal loot in any details as I did for ambu and formidon in Jimmy's post. Since I'm interested in mean return knowing their means is enough.
However, I had to model the gaps when we come to globals. The gaps are larger and I have truncated data.

I'm not interested in the derivatives of F (F'(x) = f(x)), I'm interested in expectation of x (E(x)). However, one of the results of the new model is, that I can say where the gaps do come. If I have enough data I might redo it for normal loot, but atm I don't see any need for it.

[falkao]

to let you know. The discrepancies between Hogg, Aurli and Ambu are solved. It was not the model that had a problem but my simulator. I had one superfluous term that I removed now.

Hogg behave now as it should with 1.1HP. Ambu has 1.5 HP now and Aurli is at 1.1HP with an empty correction of .1. Does anybody know the empty rate of Hogs? I don’t hunt them very often, but It seems as it is very low.

[Masta]

Originally Posted by falkao

Having got estimates for the distribution functions

Be a little carefull here, you cannot estimate the distribution with MLE, you can guess a distribution family and then estimate the parameters, even though this is probably what you mean.

Originally Posted by falkao

A probit/logit approach is not the case here. We don't have counted data and I don't want to categorize them.

You usually use probit/logit to model discreet/multinomial choice (and conditional poisson for count data). Your outcome variable for probit/logit can be 0 for no loot 1 for loot, or 0 for no global, 1 for global, for example. A variable of interest is then a probability that outcome is 1, which is a probit/logit function of covariates, such as avatar skills, kill time, ammo used, mob type etc.

Cheers.

[Konve]

"Global/hof/ath" - why are you separating these from normal loots? A global is just a 50 PED+ loot. If Mindark set the global-limit to 100 PED to reduce spam - I doubt the loot sizes would change. A hof is just a global, the system doesn't go "let's drop a hall of fame entry, this calls for big loots" - a global will be put on the list as long as there is room/it's big enough. There have been ath loots, put both into the hof section and the ath section, with the size of a small global.

So really, I understand what you're coming at with these, but you simply can't call the global, hof and ath levels - as the type is only determined by the size of other loots, not the loot itself.

I have to say I'm feeling sceptical, looking at this thread. Sure it has some nice examples of "omg it fits!" , but what else? I mean is there any point trying to make something out of this, or rather, what are you trying to make out of this? (Sorry if I missed it somewhere, it's a long thread.)

Also, the way I see it loot as always been HP related. Damage is negligible, seeing as Tantardions used to make zero damage - they have quite a lot of HP (or by the standards of that time) - and they were heavily hunted. I can't see so many people camping them if loot was poor thanks to their lack of damage potential. Actually, it fits better with my example of a pre-determined average loot output of the mob (see below) - no armor decay, no heal cost, few lag misses (big and bulky, hardly move/ed); the ideal mob to hunt as it would save you a lot of money to hunt that mob over a mob that ran around, doing damage.

Originally Posted by falkao

that's correct. MA won't pay more out than did go in. As we have discovered, loot follows some kind of an exponential distribution. So you will have many small loots, some high, and rarely some very high ones. With such a distribution it will be very hard to break even in short time. However, one of the results is that you can break even getting globals. That's what players already know.

Breaking even doesn't have anything to do with globals, though. It's like saying you're getting occasional globals but can break even with enough of the small loots.

It's all about spending less money for your kill, and trying to accumulate marketvalue in your loots. For all we know the system is designed to make player break even at (wild example); "hp/3=average loot in PEC". So you go and kill a 900 HP mob, using 300 PEC - you break even. You miss a few times or have to heal or the mob hit your armor or the mob regained some health etc - you spend 320 PEC and you loose. You might have very high skills and HP, so you don't need to heal or wear armor and your weapon is very eco and you can spend less money then the average person - you spend 280 PEC and profit.

Originally Posted by mantra

Well, I do not understand a singe word out it, but looks very mean!

I mean.........I guess 99% of the players have no idea what it is, or how it can be useful. Or it is just me?

Would you explain the subject like....you have 1 apple and 1 banana..you know..lootpool for dummies.

Well to me it looked like 10 hp mob = 10 pec loot, then 20 hp mob has to be 20 pec loot. But reading the comments that's probably not the case, still it's what I thought it looked like.


I apologize for any confusion in this post - I have redesigned it several times and moved text, so there might be something I missed. And it's 02:30 AM.

[falkao]

Originally Posted by Masta

Be a little carefull here, you cannot estimate the distribution with MLE, you can guess a distribution family and then estimate the parameters, even though this is probably what you mean.

that's correct, the sentence should be:
Having got parameter estimates for the distribution functions

Originally Posted by Masta

You usually use probit/logit to model discreet/multinomial choice (and conditional poisson for count data). Your outcome variable for probit/logit can be 0 for no loot 1 for loot, or 0 for no global, 1 for global, for example. A variable of interest is then a probability that outcome is 1, which is a probit/logit function of covariates, such as avatar skills, kill time, ammo used, mob type etc.

Cheers.

indeed, this would be another approach to deal with the data but is less informative. It would be ideal to test covariates.

[falkao]

Originally Posted by Konve

"Global/hof/ath" - why are you separating these from normal loots? A global is just a 50 PED+ loot. If Mindark set the global-limit to 100 PED to reduce spam - I doubt the loot sizes would change. A hof is just a global, the system doesn't go "let's drop a hall of fame entry, this calls for big loots" - a global will be put on the list as long as there is room/it's big enough. There have been ath loots, put both into the hof section and the ath section, with the size of a small global.

So really, I understand what you're coming at with these, but you simply can't call the global, hof and ath levels - as the type is only determined by the size of other loots, not the loot itself.

sry Konve if you've got me worng. I did use those terms so that a normal reader gets a better understanding. It is however a trivialization and therfore not correct.

What we have are mixture distributions with an upper and lower limit. Those limits are the gaps we see in the survival function.

Take a Mob with h HP, then normal loot is limited to about 3*h*r/10 PEC. The next distribution starts at h*r PEC, with r a regain factor. (Those formulas are only valid for mobs with HP>1000).

So for an Ambu with r=1.5, you won't observe loot between
about 455 and 1500 PEC.

There are 4 main subdistributions plus some others. I named those 4 normal, global, hof and ath but as already mentioned this is not correct.

c1 loot will not leed to a global, c3 with a low probability and from C4 on you will see them rather frequently.

Originally Posted by Konve

I have to say I'm feeling sceptical, looking at this thread. Sure it has some nice examples of "omg it fits!" , but what else? I mean is there any point trying to make something out of this, or rather, what are you trying to make out of this? (Sorry if I missed it somewhere, it's a long thread.)

I was interested in the return rate i.e. loot/cost. With c0-c1 loot alone, return is limited to about 40%. To get a better return you'll need loot from higher classes. This is from my point of view one of the main results I've got till now. Trivially spoken, you need at least pedders or globals to get a return >50%.

Originally Posted by Konve

Also, the way I see it loot as always been HP related. Damage is negligible, seeing as Tantardions used to make zero damage - they have quite a lot of HP (or by the standards of that time) - and they were heavily hunted. I can't see so many people camping them if loot was poor thanks to their lack of damage potential. Actually, it fits better with my example of a pre-determined average loot output of the mob (see below) - no armor decay, no heal cost, few lag misses (big and bulky, hardly move/ed); the ideal mob to hunt as it would save you a lot of money to hunt that mob over a mob that ran around, doing damage.

The only input I need to get reliable estimates is h*r (i.e. HP plus regain, so effective dmg done). Thefore I can second your opinion. So your strategy is correct.

Originally Posted by Konve

Breaking even doesn't have anything to do with globals, though. It's like saying you're getting occasional globals but can break even with enough of the small loots.

this is not correct. You can't break even without < c3 loot, due to the upper limit of nomal loot. The system is however not necessarily aimed to break even at all. Mean max return I do estimate atm is 84%.

Originally Posted by Konve

It's all about spending less money for your kill, and trying to accumulate marketvalue in your loots. For all we know the system is designed to make player break even at (wild example); "hp/3=average loot in PEC". So you go and kill a 900 HP mob, using 300 PEC - you break even. You miss a few times or have to heal or the mob hit your armor or the mob regained some health etc - you spend 320 PEC and you loose. You might have very high skills and HP, so you don't need to heal or wear armor and your weapon is very eco and you can spend less money then the average person - you spend 280 PEC and profit.

I do estimate mean loot as 0.26*HP atm and maybe it goes up to HP/3.
To make your examaple clearer. The cost to kill a 900 HP mob without regain is in ideal circumstances (maxed weapon, no fap, no armor cost) HP/3. So you'll spend 300 PEC to kill one. If retun is also HP/3 then the system is aimed to break even. The more additional cost you have, the lower your retun will be. So ECO hunting is very important.

Originally Posted by Konve

Well to me it looked like 10 hp mob = 10 pec loot, then 20 hp mob has to be 20 pec loot. But reading the comments that's probably not the case, still it's what I thought it looked like.

Mean loot depends only on h*r. So every mob will have the same ideal return rate and therfore there is no such thing as better looting mob. What we have is a shift in classes. So one mob might trigger more globals as another, but in the end return is the same.

This is very important from my point of view, since when I started I've thought I have to find the good mobs. Taking additional costs into account, a high HP mob will be more difficult to kill, due to more dmg and time that is needed to kill it. Hence you'll have more additional costs and therefore a lower return. So it might be wise to hunt only low hp mobs. However, they won't drop items with a good markup and so you might end up with less overall return if we take the market into consideration as well.

Originally Posted by Konve

I apologize for any confusion in this post - I have redesigned it several times and moved text, so there might be something I missed. And it's 02:30 AM.

I didn't see any confusion only good questions.


Edit: One further comment:

A system with a return lower than hp/3 has to be considered as gambling, from a mathematically point of view. Markup doesn't count here except you can guarantee a minimum.

Since MA claims that EU is not gambling, then maybe HP/3 will be the final result. Also Buck is suggesting that from his data.

[falkao]

Originally Posted by Dawis

..
If loot is more than 15 times greater from Aurli then one can say Aurli devastator loots better than Argo young.
If loot is less than 15 times greater from Aurli then one can say Aurli devastator loots worse than Argo young.

Originally Posted by falkao

Your proposal seems reasonable but is unfortunately wrong. Proportionality doesn't mean that.

Please read the newly added "experimental section". There I describe that mean loot might depend linearly on hp. So we have

m Argo = a*300HP + b
m Aurli = a*2800HP + b

It seems I can now remove the constant terms that do not depend on hp from the model. So in the end it might be that your proportional assumption on hp becomes true.
i.e.
mArgo = a*300HP + b
mAurli = a*2800HP + b

b = HP*c and therefore

mArgo/mAurli = (a + c)/(a + c) *HP1/HP2 = HP1/HP2

with HP1=300HP and HP2 = 2800HP

so all is proportional.