Menu
What's New
By:
Filters

Mining loot analysis

Noodles said:
last one for now. I doublechecked my taxed/untaxed data.

My lowest taxed 1 pec tt enmatter is 48 pecs (4% tax). I don't have a lot of 1 pec tt taxed enmatter data, however. The lowest untaxed is 49 pecs.

My lowest taxed 0.01 pec tt ore (lyst of course) is 93 pecs (4% tax). Also have 94 and 96 pecs at 4% tax. The lowest untaxed lyst claim is 97 pecs (three times).

Forget about the blausarium I mentioned earlier :ahh:
Click to expand...

as far as i know the lowest possible enmatter find can be 36pec (1 acid root) and my lowest crude oil find (i remember especially this one because it came at 1km depth) was 38pec. My lowest lyst ever was 93pec too.
 
Parlog said:
i am in for creating datasets for tik400 and tk320, what exactly do you need?
Click to expand...

Steffel collected

NRF number of empties till find
SIZE
DEPTH
RES type of resource
UNITS
TAXED
tt per unit
drops till find (NRF +1)
 
Noodles said:
I am almost positive that it's number 1, but the find sizes used are not the roman numerals that we see, but instead the windows that I outlined a few posts earlier. ...
Click to expand...

that has to be discussed. Have you analyzed them per resource? If you check the table for example, I have the impression that the means per size class are the same for all resources. So it might be that they are the base of randomization.
Your windows do reflect the gaps we observe in loot. It might be appropriate to use them when we know that they are universally valid and not different between resources.

Noodles said:
Originally Posted by falkao
Enmatters with high TT do have nearly no variability in the lower size classes.
Hmm, there certainly can't be a non-integer number of units, so I don't know what you mean?
Click to expand...

as you can see from the table, there are classes with a std of 0 implying that always the same value was observed.


Noodles said:
What do you mean by "variability within size classes"?
..
I do think that for higher classes (globals and whatnot), the distribution is probably not uniform.
Click to expand...

Take a resource and check the std within each class, you'll see it is rather small but becomes larger with higher classes. It might also be that the uniform distribution within a class disappears in higher classes but I don't have enough data about that yet.


Noodles said:
..
I think it's incorrect to model things using the roman numeral size classes. Instead, use the classes that the data shows to exist.
Click to expand...

If the means per natural given size class are the same for all resources, we should use them and not invent others. You're mainly faced on the gaps in loot to derive classes. Sometimes a natural class is missing leading to a gap, sometimes the range itself from one to the next class does have a gap. Maybe in the end this comes to the same result.



I'm pretty sure MA does one of the following:

An overall payout according to hit rate is set. Size classes do have a predefined mean. Starting from this mean and using the resource TT the range of units within class is set (if a uniform distribution within class is used, otherwise I don't know yet). This leads to a new mean per resource. This means per size class should then sum up to the predefined payout and hence the weights are set accordingly.
 
tax

Here some further insights on taxation

[br]Click to enlarge[/br]

The two enmatters with freq. 29% and 42% do give an indication about taxation. There seems to be a downshift from class 4 to 3 but also an upshift from 5 to 6. Both effects are stat. significant for the 42% resource p = 0.006 and p=.009 respectively.
 
Using your approach Noodles I would model it as follows.

Fig. 1: Loot below 10 PED

[br]Click to enlarge[/br]


Fig. 2: Loot above 10 PED

[br]Click to enlarge[/br]


Loot below 10 Ped seems to have as you describe a uniform distribution. The 10 PED limit is a result of the amp and is not universally valid. But nevertheless. It splits up 50% to 50% to loot from 1.56 to 3.23 and 3.84 to 9.6. So we have the first 2 classes.

Loot above 10 PED does follow a GPD with mean 61.2. Class C3 has a re. freq. of 4.4% (from data).

Hence the loot classes might look like the following.

Code: 
	weight	cl mean	weight. m
C1	0.478	2.395	1.145
C2	0.478	6.72	3.212
C3	0.044	61.2	2.693
Total			7.050


In total we do get a mean of 7.05 PED. Applying the 28% find rate we do get a return rate of 98%. This is higher as I've previously estimated with the GPD alone (mean loot 6.44).

The problem about this approach is the estimation precision. I had only 46 datapoints to estimate mean loot above 10 PED and hence the mean might be imprecise. Furthermore, the lower and upper limits for classes C1 and C2 are set by data and are not estimated.

Nevertheless, return rate is in both cases above 90%. Noodles, what do you get from your data?

Add:
here is a simulation of the above model and the observed loot.

first part till 40 ped

[br]Click to enlarge[/br]

right tail from 10 PED on

[br]Click to enlarge[/br]
 
Last edited:
Parlog said:
as far as i know the lowest possible enmatter find can be 36pec (1 acid root) and my lowest crude oil find (i remember especially this one because it came at 1km depth) was 38pec. My lowest lyst ever was 93pec too.
Click to expand...
Interesting. Your lowest lyst is in line with expectations for a 4% taxed area. Based on my observations to date, 38 pecs of oil is not possible unless the tax rate is around 20%. But there is no reason to doubt your memorable observation, so this is a head-scratcher. My avatar is only one year old--is it possible that the mining system sometime before then gave such low loots? (I do read that there used to be size I claims). I did have a claim once that gave less than the "estimated find size." Like the deed said it was a IV and it was actually a III; maybe a rare bug?

The single acid root (which is 32 pecs) should be possible if you draw a (random) number below 64 pecs and the algorithm rounds down. According to my current thinking. That has not yet happened to me, but I only have data for about 15 unamped acid root claims.
 
falkao said:
that has to be discussed. Have you analyzed them per resource? If you check the table for example, I have the impression that the means per size class are the same for all resources. So it might be that they are the base of randomization.
Your windows do reflect the gaps we observe in loot. It might be appropriate to use them when we know that they are universally valid and not different between resources
Click to expand...

The data I have shown is a composite of six resources (oil, typo, lyta, garcen, alic, and nexus). The number claims ranges from 124 for oil to 37 for nexus. I have been hoping to compare resources to each other, but don't feel there is enough data yet. I could try to compare oil to typo and see what happens.

It will be true that the means per size class (II, III, etc.) will not be the same for all resources, though most might behave. For example, the only size II azur pearls claim will have a value of 96 pecs (1 unit). Interestingly, this value is in NONE of the loot classes I am proposing. It either results from rounding up from the smallest class, or rounding down from the second smallest. I suspect the latter, but it could be both.


falkao said:
as you can see from the table, there are classes with a std of 0 implying that always the same value was observed.
Click to expand...

Ah yes, I just described another example.

falkao said:
If the means per natural given size class are the same for all resources, we should use them and not invent others. You're mainly faced on the gaps in loot to derive classes. Sometimes a natural class is missing leading to a gap, sometimes the range itself from one to the next class does have a gap. Maybe in the end this comes to the same result.
Click to expand...

I think we will need to invent others, as the data implies. Depending on what you mean by "all resources" . . . As Anna Ufo mentioned, her unamped ore IX's are smaller than her unamped enmatter IX's (I'm assuming her observations were for unamped). Which is what my data also show. So ores and enmatters would have different means for IX size. Also see comments on azur pearls above.

falkao said:
I'm pretty sure MA does one of the following:

An overall payout according to hit rate is set. Size classes do have a predefined mean. Starting from this mean and using the resource TT the range of units within class is set (if a uniform distribution within class is used, otherwise I don't know yet). This leads to a new mean per resource. This means per size class should then sum up to the predefined payout and hence the weights are set accordingly.
Click to expand...

I think this sounds reasonable. Let's say we start with a size class that has a predefined mean and width, and a uniform distribution. Then low tt resources should show a uniform distribution within this size class. I would like to know what happens to the high tt resources--do they too have a uniform distribution, or is it skewed to certain values due to rounding or truncation?

I like your next post too, but may not be able to respond to it tonight.

Edit: realized I could study the high tt resources from steffel's data: let's see what happens . . .
 
Last edited:
Apart from saying its an awesome job you guys are doing.....I can't really find anything else to say that would fit in.

Anyway, hope I'm not bothering you guys, but, since I'm a noob miner, using TT finder, I sorta had this encounter on TI mining randomly.

I don't normally take note of this but, there is this one occasion where I found an IX claim of lytairian dust (Best claim I had, so I took extra notice).

84 of them mined, amounting to 15.96 PED (TT), but the suprising thing was the depth shown on my finder was 600+ meters deep! And I haven't even maxed my finder yet. I think my skill using the finder only gave me 160+/204 avg. depth. So this leads me to think that depth is of no importance? And its only there as a distraction. What do you think?
 
sawachika said:
...So this leads me to think that depth is of no importance? And its only there as a distraction. What do you think?
Click to expand...

from Steffel's data there seems to be no difference between enmatters and depth. He is rather skilled and hence I'm not sure if this finding holds for everyone. Since finders do have a mean depth and you need skills to use them, I'm rather sure that covering more depth will lead to a more consistent find rate. But this is not yet proved.
 
here some results using data alone.

The hit rate is estimated as 27.9% with 95% confidence interval ranging from .2657 to .2931.

Using bootstrapping I can estimate the standard error of the mean and construct a 95% confidence interval. It ranges from 6.417 to 9.6685 PED.

Combining both results we do get an estimated return rate ranging from 84.5% to 140%. As you see, although having a large dataset the estimation error is still quite large and hence we can't be quite confident about our findings.

Nevertheless, with the actual data we can be rather be sure that the min. expected return rate is higher than 84%. From the approximations we did, we find that it might be somewhere between 90-99%. From the above mentioned CI also a return rate great then 100% can't be excluded, but I'm rather sure that this, if exists, is only true for some avas or is a result from sampling.


One thing I can conclude so far, is that it seems that mining has a better return rate as hunting due to the missing additional costs like armor or fap decay.
What I can't judge is the influence of skills. Maybe you need to be skilled to get a return rate above 90%. If this is the case, then hunting and mining return would be more or less the same if the right hunting equipment is used. But let's see, maybe we do get more data about this.


Contrasted to hunting loot, Steffels mining loot seems to have a return rate of at least 60% without minis. Mean loot below 10 PED has a 95% CI ranging from 4.6405 to 4.9453 which corresponds to a return rate ranging from 61% to 71%. With hunting this seems different, depending on mob this goes from 45% to 60%.
 
Last edited:
sawachika said:
I don't normally take note of this but, there is this one occasion where I found an IX claim of lytairian dust (Best claim I had, so I took extra notice).

84 of them mined, amounting to 15.96 PED (TT), but the suprising thing was the depth shown on my finder was 600+ meters deep! And I haven't even maxed my finder yet. I think my skill using the finder only gave me 160+/204 avg. depth. So this leads me to think that depth is of no importance? And its only there as a distraction. What do you think?
Click to expand...

falkao said:
from Steffel's data there seems to be no difference between enmatters and depth. He is rather skilled and hence I'm not sure if this finding holds for everyone. Since finders do have a mean depth and you need skills to use them, I'm rather sure that covering more depth will lead to a more consistent find rate. But this is not yet proved.
Click to expand...

Probably off-topic, and no evidence, but I always thought depth has no impact on chance of a hit, but an impact on what the resource is (ie. deeper claim, more likely to be valuable resource).
 
tax part II

sry, but this is a work in progress thread and hence there might be contradictory findings.

Till now I didn't find any evidence about taxation using the whole sample. This was in some way not very convincing. Hence I did a subgroup analysis splitting loot into the classes 0-4 PED, 4-10 PED and above 10 PED.

The 2 lower classes now look like the following:

Fig. 1: Mining loot contrasted between taxed an non taxed loot

[br]Click to enlarge[/br]

In both classes I do find a significant difference in loot with p = .018 and p = .007. The difference is mean between 4% - 5%. So we finally have a prove that taxation is applied as expected.

Why this finding in subgroups but not the whole sample? Having a heavy right tailed distribution sampling is rather problematic and hence I couldn't find any effect in the whole sample. Only with a very large sample I should be able to find the same in the whole sample.
 
Last edited:
Jimmy B said:
Probably off-topic, and no evidence, but I always thought depth has no impact on chance of a hit, but an impact on what the resource is (ie. deeper claim, more likely to be valuable resource).
Click to expand...

Also using subgroups I don't find any difference between resources (4 different types) and depth in Steffels data. So this still remains a mystery. Maybe Noodles does have something to add?
 
falkao said:
Hence the loot classes might look like the following.

Code: 
	weight	cl mean	weight. m
C1	0.478	2.395	1.145
C2	0.478	6.72	3.212
C3	0.044	61.2	2.693
Total			7.050


In total we do get a mean of 7.05 PED. Applying the 28% find rate we do get a return rate of 98%. This is higher as I've previously estimated with the GPD alone (mean loot 6.44).

Nevertheless, return rate is in both cases above 90%. Noodles, what do you get from your data?

Add:
here is a simulation of the above model and the observed loot.

first part till 40 ped

[br]Click to enlarge[/br]

right tail from 10 PED on

[br]Click to enlarge[/br]
Click to expand...

Awesome! I love this first simulation. I think you should model C3 and C4 (centered at ~25 and ~68 peds respectively) using the same approach as C1 and C2. My guess is that C5 is where the nonuniform distribution(s) start(s).

What do I get for total return? I don't know yet, because I have little data for C5 and above. But for unamped enmatter, with the following for C1-C4:
Code: 
Class	mean (pec)	frequency
C1	   63	          0.48
C2	   177	          0.48
C3	   634	          0.02
C4	   1730	          0.01

I get about 70% return from these classes assuming 28% hitrate. The rest of the return comes from C5 and above.

EDIT: I erred in the return calculation. C1-C4 give 79% return according to my data. Whoops . . .

Your means for C1-C4 should be 4x as much as mine. I notice that your C1 and C2 means are a bit lower. Some of that difference might be that my loots are tax-corrected, don't know about yours. Or maybe just randomness and finite dataset.

Found some interesting things when looking at steffel's solis/azur results, will work on when not holding a sick child.
 
Last edited:
falkao said:
Also using subgroups I don't find any difference between resources (4 different types) and depth in Steffels data. So this still remains a mystery. Maybe Noodles does have something to add?
Click to expand...

I didn't record the depths, so no comments. But Immortal is keeping a great thread of depth as a function of resource, and resources found using different finders.

Mining Depths Ores Enmatters Finders
 
falkao said:
[br]Click to enlarge[/br]
Click to expand...

This table is quite telling I think. The first resource has a tt of 78 pecs/unit and the second has a tt of 96 pecs.

So there are two large tt resources. The size of one unit puts constraints on how many pecs a claim can be worth (multiples of unit tt value).

Let's consider the model I've been working with, namely classes with uniform distributions. For 104-amped enmatter, the first three classes correspond to (approximately)
C1 1.96 to 3.24 peds (~48% frequency)
C2 4.72 to 10.00 peds (~48% frequency)
C3 22.8 to 28.4 peds (~2% frequency)

These ranges should shift down a bit if taxed. But we will ignore tax for now.

An obvious question immediately arises: How can you get 3.84 peds of a tt=96 pec resource (the only claim size for size V in Steffel's data), which is not inside any of the proposed class ranges?

Answer: (here's a scenario) Two possible claim sizes for this resource are 2.88 peds (3 units) and 3.84 peds (4 units). Assume the loot algorithm randomly assigns a class and value within the class to a particular claim. What happens if the loot algorithm assigns C1, 3.02 peds (or any number above 2.88 peds) to the value of the claim?

Obviously, you cannot dig up 3.15 units (3.02 peds) of a resource. So you either get 3 or 4 units. If you always got 3 (2.88 peds), that would be unfair. Though it could be a hidden source of "decay" for MA, if they chose to do it that way. If you always got 4 units, then MA would presumably lose consistently.

(here's the hypothesis) To circumvent this problem, MA randomly gives 3 or 4 units, in a weighted manner. Since 3.02 peds is closer to 2.88 peds (3 units) than 3.84 peds (4 units), you are more likely to get 3 vs 4 units. For a 3.02 ped claim assignment, you then have a

(3.02-2.88)/0.96 = 14.6%

chance of getting 4 units, and an 85.4% chance of getting 3 units.

If you follow through on this hypothesis, and play around with falkao's table of steffel's data, you can reach the following comparison.
Code: 
# units	   observed	calculated
     2          38	30.5
     3          57	53.5
     4           6	  4.8
     5           6	  9.4
     6          14	16.1
     7          16	16.1
     8          21	16.1
     9-12          22	31.1
    Larger            5	7.4
     Total         185	185

The agreement is obviously not perfect, but seems reasonable given only 185 loots.
Is this hypothesis the answer? I don't know, but it's the best so far, in my opinion. It should also be noted that the simulation would give no size 12 (and few size 11) claims, but those cannot be discerned from the real data.
I will try to do a similar table for the 78 pec resource; we'll see . . .
 
Last edited:
Here is the simulation for the first resource in the table:

Code: 
#units	observed	calculated
      2	   2	    4.1
      3	   29	    29.6
      4	   18	    21.9
      5	   0	    0.4
      6	   1	    3.7
      7	   13	    8.3
    8-10	30	24.9
   11-15	14	19.3
   Larger	10	4.7
   Total	117	116.9

Again good, but not perfect agreement. The simulation gives no size 14 and 15 claims, but those cannot be discerned from the real data.
 
Last edited:
Noodles said:
... For 104-amped enmatter, the first three classes correspond to (approximately)
C1 1.96 to 3.24 peds (~48% frequency)
C2 4.72 to 10.00 peds (~48% frequency)
C3 22.8 to 28.4 peds (~2% frequency)

These ranges should shift down a bit if taxed. But we will ignore tax for now.

An obvious question immediately arises: How can you get 3.84 peds of a tt=96 pec resource (the only claim size for size V in Steffel's data), which is not inside any of the proposed class ranges?

....
Click to expand...

I do think that the classes do exist per resource. Therefore I do not see a need for a 3.84 correction. For our analysis however, we can mix them under the condition that their rel. freq remains constant. This only to have enough data to start with.

Shouldn't C1 start with 1.56?
 
Noodles, I've adjusted for tax and did 2 approaches.

1) Pragmatical approach using data only
Lower and upper limits for return rate are based on the CI of the hit rate

Code: 
	range		weight	cl. mean	weighted m
C1	1.56-3.84	0.480	2.59	1.243
C2	4.8-10.94	0.477	7.17	3.420
C3	22.45-28.35	0.026	26.44	0.687
C4	62.7-77.17	0.014	70.03	0.980
C5	>174		0.003	304.89	0.915
				
Total					7.246
Return Rate				1.003
lower					0.954
upper					1.052


2) Model: C1 and C2 do use a uniform distribution, whereas C3 is a GPD. Range was derived from data using a correction for the highest TT res.

Code: 
	range		weight	cl mean	weighted m
C1	1.56-3.33	0.480	2.445	1.174
C2	4.8-9.6		0.477	7.2	3.434
C3	>10		0.043	62.3	2.679
										
Total					7.287
Return Rate				1.009
lower					0.959
upper					1.058

For the latter model I did a further simulation (n=100k) to get a feeling about the std. error of the estimated mean. I'll get a 95% CI ranging from 7.268 to 7.575. This is still quite large, and you'll see how uncomfortable right tailed distributions are. You'll need a very large dataset to be sure and in the meanwhile MA changed something here and something there. But that's another story.

All we have till now implies a return rate ranging from 95.7% to 110%.

So in the end, both approaches do seem to work and I'm rather sure that we can conclude that Steffel's return rate is above 95%. The question is now if this finding is universally valid. Noodles, you did observe a quite lower RR?

Interestingly we still can't exlclude RR's above 100%. If this is true, then there must exist avas that are able to profit also in the longer runs.
 
Grrrrr said:
The drop rate is fairly consistent to my analysis...

1,120 bombs dropped using OF-105 with OA-103 gave me a 29.2% claim rate.
Click to expand...

thx for sharing this, do you also have data about your returns?
 
falkao said:
I do think that the classes do exist per resource. Therefore I do not see a need for a 3.84 correction. For our analysis however, we can mix them under the condition that their rel. freq remains constant. This only to have enough data to start with.

Shouldn't C1 start with 1.56?
Click to expand...

It is surely possible that different classes exist for each resource. However, it is also quite possible (and sensible) that the looting algorithm would use a single set of classes for all resources.

We have different opinions. How can we decide which is more correct? Advocating for the single set of classes, I suggest that you do two things. First, look at C1 for the two lower-tt resources you have. How does the range compare to the ~1.96 - ~3.24 ped window that I propose? Is your range within 1 unit tt of each end? Is the distribution uniform?

Second, if each resource has different class ranges, then the data for resources 1 and 2 show that the distribution within C1 is clearly nonuniform, whereas for low tt resources, the distribution appears uniform. For resource 1, range of 1.56 to 3.12 peds (2 to 4 units) (and I would argue a very small chance of a 5 unit loot, pushing the high range limit to 3.90 peds), how do you explain that 2 units was found twice, while 3 units was found 29 times and 4 units, 18 times? For resource 2, range 1.92 to 3.84 peds, 2 units was found 38 times, 3 units 57 times, and 4 units only 6 times. What type of nonuniform distribution explains these results? Why would MA use such a distribution?

So I could absolutely be wrong, but if a single algorithm explains three widely varying sets of data, then Occam might say that the single algorithm is the correct one.

The other reason I advocate for the single C1 range (sorry, a small tangent) is from observations of the way the damage stack share team option works. I've done a few small tests (thanks for your patience Donatello :) ). As we know, it is possible for someone doing less damage to get that nice single item (ESI, anyone?) in the loot. Why? My tests suggest that each player's chance of getting that ESI is related to the % damage done . . . so if I did 30% damage, and Donatello did 70% damage, then I have a 30% chance of getting the ESI. (in the test, ESI's were not used :D ) In other words, MA likely uses this type of weighted distributing in another area of the universe.
 
falkao said:
Noodles, I've adjusted for tax and did 2 approaches.

1) Pragmatical approach using data only
Lower and upper limits for return rate are based on the CI of the hit rate

...

2) Model: C1 and C2 do use a uniform distribution, whereas C3 is a GPD. Range was derived from data using a correction for the highest TT res.

...

For the latter model I did a further simulation (n=100k) to get a feeling about the std. error of the estimated mean. I'll get a 95% CI ranging from 7.268 to 7.575. This is still quite large, and you'll see how uncomfortable right tailed distributions are. You'll need a very large dataset to be sure and in the meanwhile MA changed something here and something there. But that's another story.

All we have till now implies a return rate ranging from 95.7% to 110%.

So in the end, both approaches do seem to work and I'm rather sure that we can conclude that Steffel's return rate is above 95%. The question is now if this finding is universally valid. Noodles, you did observe a quite lower RR?

Interestingly we still can't exlclude RR's above 100%. If this is true, then there must exist avas that are able to profit also in the longer runs.
Click to expand...

Nice work :) A question--what happens if you use the same C1 and C2 intervals in both models?

Why did I get a 70% RR for the first four classes? I took the data from your first model and removed the C5, and get about 87% RR for steffel from the first four classes, quite different than my results. I only weighted C3 at 2%, and C4 at 1%, so your weights are a bit higher. I don't have as much data as you do, so my weights are not as certain as yours. I don't know if that would close a 17% gap, but it helps.

EDIT: :ahh: I thought I should go back and loot at my RR numbers, and realized I switch units to peds when considering C4. So switching from 17 pecs to 1700 pecs for the C4 mean, and the same weightings, I get about 79% return . . . whoops!
 
Last edited:
Noodles said:
..
EDIT: :ahh: I thought I should go back and loot at my RR numbers, and realized I switch units to peds when considering C4. So switching from 17 pecs to 1700 pecs for the C4 mean, and the same weightings, I get about 79% return . . . whoops!
Click to expand...

Steffels return goes like this
Code: 
C1	17.2%
C2	64.5%
C3	74.0%
C4	87.5%
C5	100.2%

It's quite possible, as already seen in hunting, that with lower end gear class distribution is different. It could be that your C1-C4 is the same as Steffel's C1-C3.


Noodles said:
Nice work :) A question--what happens if you use the same C1 and C2 intervals in both models?
Click to expand...

I had to cut the 3.84 between C1 and C2 due to the simple reason that it comes only from one resource with maybe a slightly different distribution. Therefore I find a very low number of entries between 3.33 and 4.80. Since Model 2 assumes a uniform distribution within a class, letting 3.84 exist would lead to a bias.

What we do atm using the stated classes is more approach 2 than 1, i.e. we assume that there are looting classes, that might be expressed as size classes.

To decide what to do we might need some further datasets. I'm bound atm on Steffel's data that reflects one situation. We can't be sure that this is universally valid.
 
Grrrrr said:
The drop rate is fairly consistent to my analysis...

1,120 bombs dropped using OF-105 with OA-103 gave me a 29.2% claim rate.

588 bombs dropped using OF-105 with OA-102 is currently tracking at 27.2% claim rate (analysis 30% complete at this stage).
Click to expand...

indeed, with un-L finders you have this kind of consistency, so bad times are mildly better and good times are mildly more average with a OF-105 making it a longterm finder thats able to give you about 90% tt out on average. ( which is not enough anymore since the majority of ores fell deep in MU)

good times and bad times: in my point of view even 588 bombs deliver no reliable data if you drop them in a single day. There are periods you will simply loose huge with it, others that simply make you rich and on average you get kinda even out (after Mu and with skills included). This alone can be found out with long runs that take place at a single day.

streaks: there are days youre on a roll and 20 globals a day are easily possible on CND and amped (which doesnt say youre winning though). My question would be: how do you trigger such days? how to avoid bad loot-periods and how to keep the streaks rollin in. I guess these questions are the deepest reason to make any kind of loot analysis. In my point of view entirely avoiding bad loot periods is not possible, so the main question changes to: what to do, if you find out youre in a bad loot period? And it seems like "stop it" is the best answer so far.

Btw, i had to sell my tik400 cause i didnt stop it...

so TK320 only for now.
 
Parlog said:
..
streaks: there are days youre on a roll and 20 globals a day are easily possible on CND and amped (which doesnt say youre winning though). My question would be: how do you trigger such days? how to avoid bad loot-periods and how to keep the streaks rollin in. I guess these questions are the deepest reason to make any kind of loot analysis. In my point of view entirely avoiding bad loot periods is not possible, so the main question changes to: what to do, if you find out youre in a bad loot period? And it seems like "stop it" is the best answer so far.

Btw, i had to sell my tik400 cause i didnt stop it...

so TK320 only for now.
Click to expand...

Unfortunately I stopped mining 2 years ago, so I can't speak from a personal experience. All I do atm is to analyze provided data.

I'm not sure about those bad and good periods.
When analyzing Steffel's data, the waiting times till the next find are perfectly random. So no indication about a cycle.

Mining might be quite different as hunting. You might need to know the spots and if somebody else was already on the field or not. When the asteroid dropped, must be some years back, MA filled up spots with specific resources and/or changed the hit rate. At least this was one indication that they can influence it.

I think without any further data, we can't conclude anything about that.
 
Noodles said:
..C1-C4:
Code: 
Class	mean (pec)	frequency
C1	   63	          0.48
C2	   177	          0.48
C3	   634	          0.02
C4	   1730	          0.01
Click to expand...

If I compare your class means to Steffel ones, applying an amp correction of 4 we do get more or less the same means for C1 and C2 with more or less the same weights.

Noodles vs Steffel
C1 63 vs 61
C2 177 vs 180.

Furthermore, also C3 and C4 are comparable when using model 1 (data driven)

Noodles vs Steffel
C3 634 vs 661
C4 1730 vs 1750.

Since you have not that much data about C5, and I guess you've to wait for a nice hof, we can't compare that class yet.

Steffel has a return rate of 98% till C4, you mentioned 79%. This difference is higher as expected from the means and hence I guess you're weights in C3 and C4 are to low, i.e. you didn't get enough minis and globals yet ;) What is the sample size of your dataset?
 
falkao said:
Steffel has a return rate of 98% till C4, you mentioned 79%. This difference is higher as expected from the means and hence I guess you're weights in C3 and C4 are to low, i.e. you didn't get enough minis and globals yet ;) What is the sample size of your dataset?
Click to expand...

Isn't Steffel's RR to C4 87.5%?

The sample size I presented is about 370 finds among six enmatters (the six I most commonly found). I have some data from the past two weeks that needs to be added, and I could also include the data from the resources I find less frequently. The data from the past two weeks is interesting--I've had some good returns so that should bring up the average a bit. My avatar turned a year old: some nice birthday presents :) Perhaps we should study the "birthday effect?"

Also have unamped ore, 101 amped ore and enmatter to sort through. Some nice hits with the 101 amp will also boost the return in my data.

I'm glad to see that the averages worked out, I was going to make a similar post when I got home. Also, you can break down the four resources from Steffel's data; the average for each individual resource is consistent as well.

I feel like we are becoming more confident in defining classes C1-C4; nice to see some progress :wtg:
 
Noodles said:
Isn't Steffel's RR to C4 87.5%?

:
Click to expand...

oops, yes. Guess I did read wrong my own post. With 370 finds you can't expect that much from C4 and C5. In Steffels data their freq is 1.4% and .3%. Hence with 370 finds the probability to get more than 10 C4's is about 2% and more than 3 C5 is about 3%.

So you should have ideally around 5 C4's with max 10, and 1 C5 with max 3.
 
to give some insights about what we're doing.

Fig.1: Loot classes C1 and C2 as seen from the survivor function. Please note the single 3.84 dot that I've excluded due to a single res. observation (maybe due to it's high TT)

[br]Click to enlarge[/br]

Fig.2: Loot classes C4 and C5. Please note the not perfectly uniform distribution in C3 as a result of the mixture of the 4 resources.

[br]Click to enlarge[/br]

Noodles, I think when we combine yours and Steffel's dataset standardizing for amp, we should get quite a complete picture.

Furthermore I don't think we have the need to understand how MA will round to unit TT. If the intra class distribution is sufficiently represented then we already have what we need.

I did some simulations for the Weibull model as well. The results are not convincing, I'm able to approximate intraclass means but not their frequency. So it seems as we can use the same approach as with hunting loot with the only difference, that we don't need to find a relationship to hp.
 
falkao said:
Furthermore I don't think we have the need to understand how MA will round to unit TT. If the intra class distribution is sufficiently represented then we already have what we need.
Click to expand...

Lovely graphs :) I do hope that the non-uniformity in C3 disappears when more data is added.

Anyway, whether or not we "need" to understand how MA rounds to unit tt (or a number of other details) depends on the question you want answered. If you just want to know the rate of return, all you need is the mean and frequency for each class, and the average hitrate and finder/driller decay. If you want to know your chances of getting a global, you only need to know the class frequency, average hitrate, and possibly the mean and shape of the largest class(es). You could also add tax effects to these two questions.

If you want to formulate a complete picture of the loot distribution, then the details are fun to work with, and necessary for the complete picture. Additionally, understanding the details usually helps us to understand the bigger picture as well, even if the ultimate conclusions do not include the details.

*gets down off of soapbox* :D

Still working on collating the rest of the data in my spreadsheet . . .
 
You must log in or register to reply here.
Back
Top