How to analyze loot

[kalanen]

Originally Posted by falkao

quite possible, but we should try to get any data about it. How should a setup for it look like?

Tough one

Methinks, to test this properly, an uber (no chipmonster) and noobie is needed and both of them opalize berys or such for a rotation of 1 prof standing, counting nothing but no of no loots.

And now, all ubers volunteering to shoot small mobs with opalo for 1 prof standing rotation, hands up!

What!!?? Nobody?

[falkao]

Originally Posted by kalanen

..

Methinks, to test this properly, an uber (no chipmonster) and noobie is needed and both of them opalize berys or such for a rotation of 1 prof standing, counting nothing but no of no loots.

oh oh, this could take some time.

It might also make sense to not wait for the next prof. standing, but simply compare if two avas with quite different hunting skills, have different loot (value, cycle) using the same equipment on the same mob.
Value might be different intrinsically, due to less dmg done by the higher skilled one (he kills faster), but this could be corrected with number of hits.

[kalanen]

How about this one: avatar with 50+ levels in any ranged but (practically) no melee and completely fresh player do tt fisting on same small mobs for 1 whole prof level rotation (as low as possible).

I still concider 1 prof level being shortest meaningful "phase" as it seems to contain (at least so has been for me) allways same amount of ups and downs.

That shouldn't take too long...

[falkao]

here another example of an intra avatar memory.
The data was provided by Mister X, an avatar that wants to stay anonymous. It consists of 85 runs on the same mob each with about 300 kills, so its data of about 25000 kills. Per run return rate was calculated (i.e. total loot/ total cost).

Fig. Seq chart of return rate

Click to enlarge

The next statements have been falsified in the meanwhile. Thx Noodles for pointing that out

The sequence chart looks very random, besides some maybe slight upwards trend in the last half. The runs test for example gives a p-value of p = .862. Partial correlation coefficients of observed values and observed differences do not show any departure from randomness. However, the augmented Dickey-Fuller and Phillips-Perron unit roots tests do give highly significant p-Values (p<0.001). From this we can reject the null hypothesis of unit root and hence conclude that we do not observe a random walk or better a series with unit root. What both tests did find is depicted in the next figure.

Fig. Return rate at n contrasted to the difference to the next (xn+1 - xn)

Click to enlarge

Contrasting the actual return rate to the difference from the next one to the actual one (i.e. xn vs. xn+1 – xn) does show a negative correlation. Spearman’s rho is -.681 with a p value < .001, so once again highly significant. R^2 is something like .46 and hence we can explain about 46% of the observed variance by the correlation between actual return and the difference to the next. Sure, it would be possible to predict the next return rate from the actual one, but prediction would not be very precise. One principle however is rather obvious, if the actual return is high then the next one tends to be lower and vice versa. To achieve this some kind of memory is needed.

With the waiting time analysis and the actual return rate analysis of runs we have evidence about an intra avatar memory. Data is related however, to only those two avas and since we know nothing about how loot is implemented, I’m not sure if results can be generalized. So if somebody else is willing to provide return on runs, we can check if this is a general effect.

[steffel]

interesting post.

Originally Posted by falkao

we can check if this is a general effect.

any chance to make a graph excluding globals/hofs similar to the green line here Hunting statistics ... again
to see if your data is similar to the data from over there?

[falkao]

Originally Posted by steffel

..any chance to make a graph excluding globals/hofs similar to the green line here Hunting statistics ... again
to see if your data is similar to the data from over there?

For the diff analysis I couldn't exclude globals/hofs, otherwise there was the risk loose information. However, it is possible to make a buck cum return chart without them. I should have time tonight to make it.

[BlackHawk]

Why is this not sticky?
Great work. Would be nice to point out to some math books that we can study to have a better representation on what is happening. I'm sure you have a better culture in this field than most of us. I stopped studying math long time ago and never studied intensively statistics, but I would start now, since I think it would be good to understand many things in EU. And is good that we have a scientific approach of the system.

[Jimmy B]

Originally Posted by BlackHawk

Would be nice to point out to some math books that we can study to have a better representation on what is happening. I'm sure you have a better culture in this field than most of us. I stopped studying math long time ago and never studied intensively statistics, but I would start now, since I think it would be good to understand many things in EU. And is good that we have a scientific approach of the system.

I think falkao linked to some reading earlier in the thread...

[falkao]

Originally Posted by steffel

any chance to make a graph excluding globals/hofs similar to the green line here Hunting statistics ... again
to see if your data is similar to the data from over there?

here is the promised graph

[falkao]

Originally Posted by Jimmy B

I think falkao linked to some reading earlier in the thread...

yep, there was something. I'm still not sure what one might need to start with. Maybe a good textbook that introduces some methods like this one.


The problem is, that we had to use quite different methods, some of them very specialized ones. So it might be quite difficult to understand everything especially when starting with.