Dark Orbit - finally a "real world" application for math

Ah, if only I could read German! That's a lot of data, and even with some wrong information thrown in, it would seem that 0.6% is the answer. Which is consistent with my other sources.

 

Update with some real data to 1.1
Totally proves my point

And I am pretty sure now that the breakdown for Alpha/Beta/Gamma parts was changed to 33% each. This finally explains why people were complaining about gates becoming more expensive. Well Dark Orbit players are complaining all the time, but this time it was partly justified.
The average number of turns needed to complete the Alpha gate was increased by this modification.
But dont worry, the overall cost for the first 3 gates is even lower now.
I will update 1.2 as soon as I can and finally add 1.3.

 

1. Galaxy Gate Generator
1.5 What is the breakdown of parts for Alpha/Beta/Gamma gates

It is not given anywhere how the breakdown of parts for the first three gates Alpha, Beta and Gamma currently is. It used to be 45%/30%/25%, but obviously was changed some time ago. Because the breakdown is important to determine the average costs for these gates, I started this analysis.

It all came down to counting how many parts you get for which gate. Special thanks to Ђζũέ«•»Ђάяяέŧŧ™[ØĦ] for providing his sample of 275 parts. I added another sample of 700 parts, resulting in a total sample size of 975 parts.



The results are very clear.
Apparently, the breakdown is 33.3...% for each gate.
Although the uncertainty in the results would allow for a slightly different interpretation, it is way more plausible that the breakdown was set to this distribution instead of something weird like 36%/30%/34%.

Of course this change was not announced officially by BP, so one might argue that this is just another hidden rip-off in the game.
And in fact the average number of turns needed to complete the Alpha gate is higher than it was before this change. That is the reason for all the diffuse complaints about gates becoming more expensive.
But in fact, the average cost for all three gates combined is lower with the new breakdown of parts (see the table in post #17). I wonder if BP had this in mind when they changed the probabilities

 

Hi El_Burro this data is statistically similar to what I am seeing, in relation to the break down in A,B,Y distribution for each gate part yield.

This is why I suggest there is further manipulation server side. As the .php give 13% probability, so if there was not further manipulation ( I'll use adjustment just to stop side debate ) the distribution should be as whoeva and I have posted. (whoeva's post )That means when hitting a gate part there should be a (34/164x100) 21% chance it's alpha, (48/164x100) 29% chance it's beta and (82/164x100) 50% chance it is a gamma part.
So to get 33% for each gate there must be server side adjustment. or the 33% for each would be closer to 21%/29%/50% respectively. Then if you work out the reduction in cost using prem rebate, if we did indeed see a return of 21%/29%/50% A,B,Y would have a significant potential to return a uri profit.

Even with the original A,B,Y distribution returning a typical 4,2,1 distribution it was potentially possible to return a profit from doing these gates.

So Questions:-

Assuming a return of :-
A. 33%, 33%, 33%
B. 21%/29%/50%
Using prem rebate, and assuming the materialiser also returned a 13% gate part yield, with the same relative distribution.

What would be the % of the population that would make a profit from continuously doing A,B,Y gates.

^ this along with limited multiplier data (as multiplier yield requires much more total data ) is why I perversive that there is also some adjustment to the multiplier yield.

I would say that the change is possibly due to the Kronos as when DO change the typical A,B,Y relative distribution they did not change the relative requirement to get Kronos gates.

 

As I already stated, there is no right or wrong when it comes to the breakdown of gate parts. From a programmers point of view, there are simply 3 calls to the pseudo-random number generator.

1. Decide IF a gate part is produced
2. Decide for WHICH gate (thats what you call adjustment or manipulation)
3. Decide WHICH PART for the chosen gate is produced

You could aswell do it with only 2 calls omitting step 2, but that would make it hard to implement any weighting for the three gates. Could you elaborate why you constantly labour the point that there is server side adjustment for the three gates? I dont get your point.

Concerning the "assumption" that the GGG has a probability of 13% for gate parts...
I thought that I had demonstrated once and for all that the probability is 13% by adding some real data to post #41.
Concerning the "adjustment" on the multipliers: there is none. See the data I added to post #2. Among the 1624 gate parts counted were LOTS of multipliers. If there was any adjustment here it would have shown up in the analysis.
Unless someone comes up with a better analysis proving me wrong, I can not take any of these statements seriously.
And as one of the trolls pointed out, I find it hard to believe that the guys who failed to calculate the bonuses for the item-upgrades correctly in the first place, the same guys who brought us the option to give names to the ship hangars, the guys who brought us the "AI" of the PET collecting boxes, are capable of such a manipulation. Especially a manipulation that goes unnoticed in statistical tests.

I have to admit that this is not the first time I do such an analysis. This is why I know for a fact that it was never possible to make a profit from the GGG with the old breakdown of gate parts (45%/30%/25%).
The costs per Gate Unit are as follows (including rewards for NPC, 70 Uridium per turn, cheapest strategy):

33%/33%/33% -> 4052 Uridium
45%/30%/25% -> 4892 Uridium
21%/29%/50% -> 3422 Uridium

And please dont get me started on Kronos gate
The numbers needed for each gate make no sense at all. 4 Epsilon gates now that most people going for Kronos gates dont need any more LF4?

Edit: Maybe the best argument I have that there is nothing suspicious going on with the GGG: complete a few gates and see how the numbers of turns needed compare to the values in the table in post #5. Every time I do this I am well within the expected range.

 

1. Galaxy Gate Generator
1.3 Average cost for each gate

Based on the average number of turns needed to complete a gate and the Uridium reward from the gate (including Uridium from NPC) the average costs for the gates are shown in the following table. The cost given in the table is the amount of Uridium needed to complete the gate minus the Uridium you get as a reward and from the NPC.
Remember: since Alpha, Beta and Gamma gates are linked, it makes no sense to give individual costs for each of them. Playing only one or two of these gates while letting the others completed in the generator is a total waste of ingame currency. Instead, they are represented by a "Gate Unit" GU, which corresponds to the reward you get from an Alpha gate. The rewards from the Beta and Gamma gate are two times and three times as high than from the Alpha gate respectively. Consequently, a completed Alpha gate counts as 1GU, a Beta gate as 2GU and a Gamma gate as 3GU.
So getting 1 GU from the GGG corresponds to 20000 UCB-100 ammunition, 100000 Honour points and so on.



Two strategies are compared in this table: The cheapest using multipliers instantaneously on level 2 and the fastest but most expensive one where multipliers are used automatically on level 6. Keep in mind that when you do more than one turn at a time the most expensive strategy is unsed. You might get to activate a multiplier on lower levels a few times, but you will still pay more than necessary using more than one turn at a time.

See how there is a huge difference between the two strategies especially for the Alpha/Beta/Gamma gates. You pay more than twice as much for these gates using the worst strategy with a discount of 30%.
Not to mention the difference between the different levels of discount. So whenever you do a few gates in a row, better get the discount package first.


This was more relevant for players building gates wit Uridium. For those using extra energies the yield per turn might be more important, especially the Uridium you get from it. The highest values in each category are underlined and framed for clarity.



Some of the information I found about the XP and honour rewards from NPC in the gates was inconsistent, but the differences were rather small.
I will leave the interpretation up to you because I am pretty stoked by the tremendous amount of information that converges in this table

For Comparison, this is what you get on average from 70 000 Uridium with 30% discount:

 

Just wanted to say, I think this is turning into the most useful player-made thread on the forums right now. Brilliant for players looking to get the most out of their uridium and helping people understand the generator better, great work

 

It's because this relies on Real-world math instead of assumptions.

 

Hi El_Burro here is the data I collected, shown in a similar format for A,B,Y gates. I actually stopped doing them for a while as they were consistently returning below average yield. As you can see the latest data is returning result that would be expected from a 13% yield;

However I am more currently seeing the yield for materialiser x2 gate part yield returning a mean of 8.89% for all gates, accepting that this is only based on 124 x2 materialiser results. Which is significant in the average cost of a gate especial A,B,Y gates.

I started collecting this data just over 1 year ago and have not used the materialiser recently. Also I'm not able to determine when the A,B,Y(1) data ended and the A,B,Y(2) data started, as unfortunately the data was for every energiser use so did not log dates.

I like you disregard anecdotal information and have recorded the result from every 100 spins to obtain factual data, it would have been impractical to screen print all the spin results.



From the limited Materialiser data I have, it would seem that the materialise yield is not return the expected 13% yield. I could post a chat of the Materialiser yield if really needed. Noting I have not used the GGG for a while.

I have no reason to manipulate the data and can show links to historic posts where I posted remarkable accurate yield for all items previously in the old forum.

 

Is this based on enough spins to come to a decisive conclusion? 124 x2 spins doesn't seem like very much.

 

Considering I use to make £ multimillion decisions based on the result from much less data, I am reasonable confidant that the materialiser is not returning for me at least the expected 13%. I could have been extremely unlucky, but the data would lean towards a mean yield well below 13%.

 

But El_Burro said in one of his posts that you need to spin an infinite amount of times to get the 13% gate part yield.

 

The point I labour is that it is possible to affect gate part yield server side.

Accepting my data posted above is accurate which I was meticulous in collecting. It would seem that the earlier set of data yield I obtain, is not consistent with the .php code. of 13% yield.

If there was not cell 2 (as you describe it) ( further server side adjustment) we should see an A,B,Y yield distribution closer to 21%/29%/50% respectively. But I like you have been seeing closer to 33% for each A,B,Y gate group.

 

some great useful data in this post very helpful for the whole community but again the odd rubbish being talked by some :/

 

,

When one evaluates data, the more data available will more likely return a more reliably accurate result, however it is possible to determine if even a single data result is within a normal distribution. 124 result returning a yield of 9% against a specified 13% yield could be just extremely bad luck, but looking at each result in relation to what would be expected, suggest that there is an issue with the yield for materialiser use ( for me at least).

Statically it would be possible to return extreme yield results from only using 2 materialisers spins. you could failed both time you would have 100% fail rate or win both times with 100% success, so the more data you have as long as you know the data is within the same population then it is likely to return a more accurate result.

124 results for a yield probability of 13% is not a lot of data as I first pointed out. But it is enough for me to suggest that it is significant.

If you look at my chart above, there is only just over 100 data point with only 21 data points in the second set of data. But it clearly shows to separate distribution.

 

-Pacman2- let me try to understand first what exactly we are talking about. Remember, english is not my first language so please try to be as precise as possible.

A few questions on this post:

Are these results from doing 100 spins at a time or did you do 100 single spins for every data point in the chart?
What exactly is the difference between A,B,Y(1) and A,B,Y(2)? Did you do anything different or is it just that there is some time period between these 2 samples?
When you are referring to "the yield for materialiser x2 gate part yield" what exactly does this mean? Same with "based on 124 x2 materialiser results", please elaborate what a x2 materializer result is.

Which brings us to the second post:

I guess with the answers to the questions above I should be able to comment on this.
Just a hint: we dont have to rely on "guessing" the significance of statistical data. One can usually estimate the amount uncertainty in the results quantitatively. I did this for example by adding a confidence interval to the data sampled in post #2.
This way it was possible to tell that although I got less gate parts with an interval of 1 than with an interval of 2, the data still confirms the distribution I predicted.

 

Each data point is the number of parts ( new or duplicate) for each consecutive 100 energy boosts, which would have been made up of 5 spin and 1 spin energiser uses.
However it should not make any difference if they were from 1 x 100 spins 10 x 10 spins 20 x 5 spins or 100 x 1 spins. The results are the number of parts (new or duplicate ) from each consecutive 100 energy boosts.

The only difference in the two sets of data is time with a break not doing A,B,Y gates in between the two sets of data, with A,B,Y(2) being a more recent period than A,B,Y(1)

I use the multiplier ( I often call it materializer in error my dyslexia kicking in ) on x2 and log if I get a part or not from using the multiplier. The multiplier yield is the percentage of times a part is received from using the multiplier.

The 124 is the number of multipliers the part yield was calculated from.

 

Let me start with the last issue: you were successful in only 8.99% of 124 cases using a multiplier instantly. Theory suggests that the success rate should be 13% for this experiment.

Let me estimate the statistical significance of your result:



The chart is valid for the whole range of intervals, but we only have to look at an interval of 1. Theory suggests that this event (interval of 1 between gate parts -> success if a multiplier was used) has a probability of 13%. But looking at the confidence intervals for a sample size of 124, one can clearly see that 8.99% is well within the expected range of the results.
So from the data you sampled one can not conclude that there is something wrong.
The result is not statistically significant.


Your other results with the 100 consecutive spins has a total sample size of over 10000 for the A,B,C(1) sample. If the mean value is 11%, then this result IS statistically significant. But I cant really tell what went wrong.
From all the data I sampled recently i can only conclude that the probability for gate parts is 13%.
From all the data I sampled before, starting a few years ago, I can only draw the same conclusion: the probability for gate parts is 13%.

Since at least your more recent sample A,B,C(2) appears to be drawn from a probability of 13% for gate parts, I would conclude that at least NOW the probability for gate parts is 13%. I dont know went wrong with the first sample A,B,C(1). Maybe there was a bug that in fact lead to a lower probability for gate parts. But if it was a bug, then it has been fixed before you took your second sample A,B,C(2) and before I sampled the data presented in this thread.

 

Think again. The chart contains exactly what we are talking about.

Your experiment was: You got a multiplier, now you activate it instantly and use it on the next turn. So if the interval between 2 conseutive gate parts is 1, then 2 new parts are created an your experiment was successful.
It the interval is anything but 1, your experiment was not successul.

In order estimate the uncertainty range for your experiment, it is enough to have a look at my chart at an interval of 1, ignoring the rest of it.

 

If you think my chart describes something completely different (which it does not, I am pretty sure about this) then please estimate a confidence interval for your experiment yourself.
You will see that the result you got (8.99% success instead of 13% success) is well within the estimated range of the results due to the small sample size.