Pages: 1 2 [3] :: one page |
|
Author |
Thread Statistics | Show CCP posts - 0 post(s) |
Roche Pso
Gallente Deltole Research Labs
|
Posted - 2007.09.06 03:19:00 -
[61]
Originally by: Etho Demerzel
I presented it as a simple solution to the problem. As the complex solution is too complicated for most people. In the same way ballistic movement is modeled in all high school and even in college physics as describing a parabola when the correct approach is an ellipse. Or in the same way you teach classic mechanics without considering most viscous laws, relativity and other minor effects.
Albeit inexact the results it gives are reasonably near the ones obtained by the brute force approach of Birfon Ell, as he stated in his post. Yours on the other hand is just wrong and gets considerably wronger as the number of elements in the streak grows ad the chances decrease.
I am an engineer as well, just to let you now. Good engineers know a lot more about math and models than you are showing here, and they also tend to talk about what they know,not about what they don't know.
Out of interest, are you Steve Firth irl?
|
Etho Demerzel
Gallente Holy Clan of the Cone
|
Posted - 2007.09.06 03:27:00 -
[62]
I would give you a retort for this if I had a clue of whom you are talkign about. Unfortunately, as I do not, your joke is lost on me.
=====
"If a member of the EVE community finds he or she cannot accept our current level of transparency, we bid you good luck in finding a company that meets your needs." - CCP kieron... |
Roche Pso
Gallente Deltole Research Labs
|
Posted - 2007.09.06 03:37:00 -
[63]
Originally by: Etho Demerzel I would give you a retort for this if I had a clue of whom you are talkign about. Unfortunately, as I do not, your joke is lost on me.
Not a joke, was a serious question. The tone of your posts reminds me of him. I will download LyME to my PDA and figure out how to use it. Dont have access to MathCad at the moment cos I am working off site
|
Tunak
|
Posted - 2007.09.06 18:43:00 -
[64]
Originally by: Etho Demerzel I am an engineer as well
This explains a lot.
You started this whole thread over what is basically 1.85 missions.
For a 95/5 split with the number of missions you quote you should see 77.15. You saw 79.
You obviously like finding fault with others. It's the only thing you've done in this thread. Fault with CCP. Fault with every responder. You participate in personal that have no point in a discussion.
The straight up split didn't provide enough fault so you had to go digging and find the streaks.
I don't contest that the streaks are highly unlikely to happen. However answer me this. What says your 1543 sample set has to have a perfect 95/5 split? What says your sample set has to have the perfect number and length of streaks?
The streaks did happen. Probability is not prophecy. I'm sorry you had to do/turn down one extra kill mission.
|
Heikki
Gallente Wreckless Abandon Triumvirate.
|
Posted - 2007.09.06 20:10:00 -
[65]
Originally by: Roche Pso Very quick maths giving a wrong but not too far out answer: 0.5^8 = 0.00390625 chance of getting streak of 8 1500-7 = 1493 sets of eight tosses 0.00390625 * 1493 = 5.83203125
Sounds pretty correct (assuming we accept streaks of both heads and tails): "On round 3000000 avg 5.86 streaks of 8 in 1500 tosses."
Originally by: Tunak What says your 1543 sample set has to have a perfect 95/5 split?
Don't let the failures in communication hide the message; the OP does have a valid point. If you don't like the way he presents himself, just find the gems and focus on them.
The data seems to strongly indicate (for mission distribution) that the formula is something else than we (as community) had assumed. It's not some kind of fault of CCP; it is merely correction to our lack of knowledge.
For missions it's no big deal; at most a minor annoyance.
But for invention and business like that, it might literally be question worth of billions. Something worth of exploring further.
I don't think there is much point to try to get the exact numbers or exact formulas: rather would be nice if some others involved in the areas would pay attention (==write down) the patterns they see.
Like, how many invention fails/succeeds you get, and if they depend on way you accept the jobs, or on last successes, or on phase of moon, or..
-Lasse not really involved in the said fields
|
Larshus Magrus
Elite Storm Enterprises Storm Armada
|
Posted - 2007.09.06 20:39:00 -
[66]
I know years back Everquest had a major issue with their RNG. The same thing happened as is being reported here... only for such things as rare monster spawns (locality of streaks) rare item drops, crafting skills.
The devs were using stock RNG generators provided by the OS (if I remember correctly). No one ever tested the symetery of the distribution patterns. If someone feels like it google the web and maybe dig up some of the old discussions... they were long and heated.
I'm guessing the same thing is happening here. I too have noticed rediculously low probability streaks in short periods of time. Everquest, to clear the air, released the old RNG and provided the formulae for the new one to clear the air and get player approval.
I'm suggesting that the devs do just this. If CCP is using the psudo-random (highly unlikely) or even the "random" generator in windows, this could very well explain what has been happening. The RNG windows provides is fundamentally flawed.
|
Setsunai
Eve University Ivy League
|
Posted - 2007.09.06 23:00:00 -
[67]
Originally by: Tunak
Originally by: Etho Demerzel I am an engineer as well
This explains a lot.
You started this whole thread over what is basically 1.85 missions.
For a 95/5 split with the number of missions you quote you should see 77.15. You saw 79.
You obviously like finding fault with others. It's the only thing you've done in this thread. Fault with CCP. Fault with every responder. You participate in personal that have no point in a discussion.
The straight up split didn't provide enough fault so you had to go digging and find the streaks.
I don't contest that the streaks are highly unlikely to happen. However answer me this. What says your 1543 sample set has to have a perfect 95/5 split? What says your sample set has to have the perfect number and length of streaks?
The streaks did happen. Probability is not prophecy. I'm sorry you had to do/turn down one extra kill mission.
I don't think the OP ever argued that the sample set didn't or wouldn't approach the expected values, because it did. He even said so himself. The problem is that he saw streaks that have a very unlikely (ridiculously low) chance to happen. Several times. The problem is not whether the algorithm used to determine success approaches the expected values over time or not, because it does, it's that from what has been seen its implementation is far from truly random (which is arguably not feasible), or even pseudo-random. It might even have memory just to approach the expected values. That's the problem the OP has stated.
|
Blih Nox
|
Posted - 2007.09.07 14:13:00 -
[68]
Me tries to summon Dev response with ze sekrit code [iuh%duy%&9]
|
Pwett
Minmatar QUANT Corp.
|
Posted - 2007.09.08 04:22:00 -
[69]
I've set up a mission tracking database if you guys want to start storing mission distribution statistics. See this post:
http://oldforums.eveonline.com/?a=topic&threadID=590534 _______________ Pwett CEO and Founder [QTC]QUANT Corp.
|
Nyphur
Pillowsoft Total Comfort
|
Posted - 2007.09.08 05:00:00 -
[70]
Etho, you are a mathemagician.
Press this issue in petitions in-game or direct mails to CCP and you'll likely get some kind of response. I'm sure CCP can brute force a million invention attempts on the test server to provide a sample group so large as to be irrefutable.
Eve-Tanking.com - We're sorry, something happened. |
|
El Yatta
Mercenary Forces Exquisite Malevolence
|
Posted - 2007.09.08 10:55:00 -
[71]
Originally by: Nyphur Etho, you are a mathemagician.
Press this issue in petitions in-game or direct mails to CCP and you'll likely get some kind of response. I'm sure CCP can brute force a million invention attempts on the test server to provide a sample group so large as to be irrefutable.
To be honest, I think you, and the OP, have missed the point a bit. I have no way of proving or refuting the OPs maths, and have no wish to. He may very well be correct. However, its pretty pointless petitioning CCP and bug reporting - what incentive is there to change this? Lets look at the three biggest GAME mechanics with supposed randomness:
1) Missions. Do they need to be random? Even if you have proved they are not, would it significantly improve most people's game time to get "truly random" distribution of missions? (Incidentally this is why I dont understand ANYONE doing any mission, ever, because its just a grind of repetitive tasks, where people know "I can do this mission in 14 minutes, but I decline that one because I know what it entails") Getting the same mission 5 times in a row strikes me as EQUALLY boring as getting 5 different ones that you've done before and know how to run.
2) Invention. Is it a bad thing if this isnt "truly random"? You've already said your results tend towards the expected distribution long-term, which is already a more certain money-spinning business than real life R&D. When invention gets streaky, firstly it gets rid of low-budget inventors, so keeps the field from being completely flooded, instead of just quite busy. When you're on a good streak however, you do get quite excited! In short, it adds a bit of excitement to the process, a bit of tension.
3) ECM. The suggestion that this is "more streaky" or "even less random" than missions or invention is a bit silly, because its skewed by perception. Most fights last around 30-120 seconds, and in that time you can get only 1.5-6 cycles in on your jammers - so most fights you will "feel" (in my extensive experience) like its either all gone wrong, or its all gone right. Similarly, if you are being jammed, in a similar fight, losing all your target locks for 20s+locktime is so debilitating that if you get jammed 2-3 times in a row, it will completely swing the fight against you, as well, as instilling the sense of "random being bugged" because "how else" could they have jammed you for an entire fight?
One of your two rotating signatures exceeds the maximum allowed filesize of 24000 bytes - Devil ([email protected]) |
Corporati Capitalis
Tollan Technologies
|
Posted - 2007.09.08 14:39:00 -
[72]
Originally by: Etho Demerzel So for a binary example like ours (success/failure) the chance of failure of 8 attempts in a row, supposing success ratio is 50% is (0.5)^8 = 0.39%, which albeit possible is very unlikely. The chances of 16 attempts failing is: 0.0015%
Say you've had 7 failed attempts in a row and you try again - what is the chance of THIS attempt failing? Is it 0.39%? Or is it 50%?
You show decent math skills, but that doesn't make your conclusions right.
|
Nyphur
Pillowsoft Total Comfort
|
Posted - 2007.09.08 16:24:00 -
[73]
Originally by: El Yatta
To be honest, I think you, and the OP, have missed the point a bit. However, its pretty pointless petitioning CCP and bug reporting - what incentive is there to change this?
From your three examples, it's clear that you yourself may have missed the point. Missions being random isn't particularly important and if CCP's randomising code is as old as the mission system and is working the same way now as it did then, clearly nobody has really cared about getting too many kill/courier missions in a row and it's not important. However, your other examples of where random factors are used in eve and others you didn't give as examples ARE important.
Your point on Invention is a good one but ignores the problem with an imperfect random system. If it's not random, there's a pattern to it and if that pattern can be discerned, it can be abused. Completely hypothetically, what would you do if you found out that all cap recharger II invention completion on even-numbered minutes succeeded and all completion on odd-numbered minutes failed? You'd complete only on even-numbered minutes and never fail. Or what if you found out that a new "random" number is only selected a maximum of every 5 minutes? that would mean that if you got a success, you could just repeatedly turn in all your invention jobs of the same type and get successes on all of them until the number switched.
Now what happens if ECM, wrecking hits etc aren't perfectly random? If there's a pattern to to those, some poor sod loses a ship he shouldn't have. Even if the pattern is something players can't intentionally manipulate, remember that the client can be hacked and made to run any number of python scripts that it shouldn't be and that not all players are upstanding citizens like you and I.
Those are completely ridiculous examples to demonstrate the point but the point is valid. If the system isn't performing randomly, it's performing with some kind of pattern or preference. No matter if it evens out to being random in the long-term, highly improbable streaks should not be occuring with the frequency they are. You'd expect maybe one in the entirety of eve's playerbase in a month, not two in one character in a month.
At the very least, we can conclude that something appears to be wrong and MAY be wrong and that it's worthwhile for developers to check it out. Developers have access to information and resources we don't and aren't constrained to statistics, guessing and arguing about the subject over the forum. Even if they waste a day investigating and decide nothing's wrong, it should be investigated at the very least because huge streaks of invention failures are a real and pressing issue that every big inventor experiences.
Eve-Tanking.com - We're sorry, something happened. |
Nyphur
Pillowsoft Total Comfort
|
Posted - 2007.09.08 16:37:00 -
[74]
Originally by: Corporati Capitalis
Originally by: Etho Demerzel So for a binary example like ours (success/failure) the chance of failure of 8 attempts in a row, supposing success ratio is 50% is (0.5)^8 = 0.39%, which albeit possible is very unlikely. The chances of 16 attempts failing is: 0.0015%
Say you've had 7 failed attempts in a row and you try again - what is the chance of THIS attempt failing? Is it 0.39%? Or is it 50%?
You show decent math skills, but that doesn't make your conclusions right.
That's not how probabilities work. When measuring the probability of a singular cointoss event, the probability is 50%. When measuring the probability of two separate cointoss events, each one is 50% but the probability of two IN A ROW being heads is 25%. This is something that is proven, it's a fact of mathematics. If you toss coins repeatedly, the chance of getting a streak of 8 heads is roughly 0.39%.
Saying that showing decent maths skills doesn't make him right is silly, he isn't asking you to take his word for it because of his pretty maths skills, he's presenting maths to represent what he's talking about. You're inferring that he is trying to baffle you into believing that he must be right and that simply isn't the case.
Eve-Tanking.com - We're sorry, something happened. |
Roche Pso
Gallente Deltole Research Labs
|
Posted - 2007.09.08 18:51:00 -
[75]
For what it's worth, i tossed a coin 1500 times on Friday, noting the result of each toss. Then I did the same another 599 times. After that I got bored and analysed the results.
My conclusions are that although 0.5^8 feels like a small number, rather than around a 70% chance of seeing a streak of 8 heads in 1500 tosses as Ethos' maths gives, my sample suggests that actually there is a ~95% chance of seeing it. The results would make me expect to see a streak of 8 heads occur 2 or 3 times in 1500 tosses, 6 would occur about 2.5% of the time. Of course, as Ethos' number is the lower limit, this means my number is agreeing with his.
I then did another 2 batches of 100 x 1500 tosses and in each of those two batches I got a streak of 11 in one of the samples, thats an occurence in 1% of the runs. 0.5^11 is quite a bit smaller still and maybe some people would like to argue that you would not expect to see it in 1% of the cases, but that is how it came out in my (small) experiment.
|
Roche Pso
Gallente Deltole Research Labs
|
Posted - 2007.09.08 19:02:00 -
[76]
Originally by: Etho Demerzel Now the data:
Marketing missions executed: 1543 Number of Kill missions: 79 Number of courier missions: 1464 Percentage of kill missions: 5.12% Percentage of courier missions: 94.88%
Most anomalous streaks:
1) 8 kill missions in a row Chance: 0.00000000390625% Frequence: 2x
2) 5 kill missions in a row Chance: 0.00003125% Frequence: 5x 3) 96 courier missions in a row Chance: 0.73% Frequence: 1x
Item invention (all the same item, skills and conditions)
Attempts: 900 Successes: 467 Failures: 433 percentages of successes: 51.9% percentages of failures: 48.1%
Most anomalous streaks:
1) 16 failures in a row Chance: 0.0000439% Frequence: 3x
2) 8 failures in a row Chance: 0.287% Frequence: 5x
3) 6 successes in a row Chance: 1.95% Frequence: 8x
Having done my simulations of coin tossing, and seeing the results. I dont think your results seem odd. They would be if you had another 599 showing the same thing, but you only have 1 sample, and your results fall in a range that seems to be possible.
|
Nyphur
Pillowsoft Total Comfort
|
Posted - 2007.09.08 23:57:00 -
[77]
Originally by: Roche Pso
Originally by: Etho Demerzel
Most anomalous streaks:
1) 16 failures in a row Chance: 0.0000439% Frequence: 3x
2) 8 failures in a row Chance: 0.287% Frequence: 5x
3) 6 successes in a row Chance: 1.95% Frequence: 8x
Having done my simulations of coin tossing, and seeing the results. I dont think your results seem odd. They would be if you had another 599 showing the same thing, but you only have 1 sample, and your results fall in a range that seems to be possible.
Just off the top of my head, as I'm not a mathemagician myself (god I love that word), there is an easy way to calculate the sample size in which you'd expect to see a given streak, isn't there? Like if the streak has a probability of 1%, you'd expect to see it about 95% of the time in a sample size of... what? I only ask because it's important to establish that a streak of 16 invention failures is statistically insignificant before three occurances of it out of a sample size of 900 can be ignored.
And isn't this just another argument for the devs to run a brute force attempt with a huge sample base just to disprove it?
Eve-Tanking.com - We're sorry, something happened. |
Torrilin
|
Posted - 2007.09.09 01:55:00 -
[78]
Quote: The chances of a marketing agent (95% courier / 5% kill) to offer you 10 kill missions in a row is in the order of 10^-13
Looks about right, except for one tiny detail. If you record your kill/courier ratio for a marketing agent, the ratio approaches 50/50. I won't say it is definitely 50/50, my sample size is too small (around 40 missions). There are several different lists of agent mission ratios out there, and they are not all correct. Marketing agents are a common area where lists disagree. They're by no means the only one.
If you wish to make statistical statements about a particular agent category, you can use a mission ratio list as a guideline, but you can't take it as gospel truth. Always do a verification sample yourself, and a literature check. The same thing goes for doing statistical analysis on any other "random" game event.
|
|
|
|
Pages: 1 2 [3] :: one page |
First page | Previous page | Next page | Last page |