The rng is flawed because “clustering” (a term which is difficult to define) occurs more then you would expect.
There has been a bit more indebt discussion about this recently with yellovsnow and i think also you? (though not sure of the later).
I tried to proof this with the example of 3xshifter in a row from eudora and then i got hit with “sample seize to small”.
And that is sort of the end of the road for me since i dont have the data of a very large sample seize. My example gave reason for suspicion to me at least,based on standard statistical norms with usually is 1xsd or 2xsd. And this example is far beyond those limits.
But i now see i went about this the wrong way. I should not have looked for confirmation but instead i should have looked for proof that my idea is false. To do so i would need the outcome of thousends of consecutive eudora rolls in various lobbys,and since i dont have acces to that i cant disproof it.
I have a sample seize of 5 and in that sample seize of 5 a clusering of 3 times the same outcome did occur with odds of about 1/100k. Maybe that is what you would expect from a sample seize of 5 but i certainly didnt expect that.
Blizzard no doubt fixed that by now so i am not sure its still fruitfull to look for more confirmation in recent data.
You can perform a randomness test on the outputs, but you would need to think really hard about how you are capturing data.
My involvment in that discussion was from a computer science perspective, as the source post was someone running an experiment on an rng function that returned the same numeric value when the function was called rapidly (were talking sub nanoseconds). I was just pointing out that that particular scenario is completely unrelated to seeing rune of the archmage a few times in a 10 minute span.
Clustering is generally expected in random distributions.
I could flip a fair coin 5 times and get 5 heads as a result.
The previous instance has no impact on the next instance! Thats the concept you, CONTINUALLY, fail to grasp! If it did then that, kind sir, would be RIGGING!
You could probably use the Mulligan to design a randomness test. Create a deck of single cards, queue into a thousand or so games, and see what your first cards plus full Mulligan are. Then see if the distribution is normal.
Ok but what about a 50 sided coin flipped 5 times and in that 5 times you see 50 show up 3 times in a row.
Thats the correct comparison. Arguably still a small sample seize but the event is rare enough to raise suspicions. There has been many posts dedicated to this example already so i am not going into that again.
Eitherway its bedtime for me anyway so good night to all of you. I will come back to the discussion tomorrow after i had a good night of sleep and my head is clear again.
Your sample size is not 3. 3 is the sample size you’ve seen! The universe of possibilities is every eldora drop.
This is like you saying every cat is white because of all the hundred of millions of cats, the three in your street in particular or the three you know are all white.
It’s not an impossibility, you’re just looking at a very small subset of cases where the improbability happens. You may as well say 3 eldora drops generated shifters, and that 100% of eldoras generate Shifters because that’s the part of the universe you’re focusing on and therefore it’s rigged.
That’s something the simulation described in the opening post calls into serious doubt. So I’ll disagree here.
When you win a game and then immediately requeue, your more likely to see either the same deck or a deck that beats the one you just faced. When you lose a game, you’re more likely to see either the same deck or a deck that loses to the one you just faced.
No, not so long as the instance-to-instance impact is matchmaking by winrate working as advertised. Which it is.
You can disagree all you want. Run your simulation 1000 times, not just 5 times. % times is also not worth it takes hundreds, if not thousands, of parses to get a proper analysis.
It doesn’t really matter how many parses, the previous instance is literally an input, and part of the process and so it affects the output. That’s not the same as saying previous chance based events affect the next independent chance based event.
His experiment drastically magnifies the sorting effect of the algorithm though, and so the effect would be far less significant (but still exists) in an MMR system.
and also to what it’s modeling. After you win a game, your MMR/rank is one win higher than it was one game ago. Which means if you immediately requeue, you’re going to have a significant chance (not an RNG chance but a time in queue pairing) of facing an opponent who was at your rank one game ago, won, and immediately requeued.
Assuming you mean in hearthstone/equiv, there’s an assumption there that the at-your-rank-one-game-ago player is playing. But I understand your broader meaning.
If you mean "sorting effect per round," then I heartily agree. Good and bad matchups in the simulation are polarized at 0-100, in Hearthstone it’s not so one-sided.
If you mean overall, I’m not so sure. How many games does a typical Hearthstone player play per month? Probably more than 64. Perhaps a lot more. And even if I’m exaggerating the sorting effect per round by a factor of 5, what if there are ten times as many games played?
True. But I’d say it’s an assumption that’s correct more often than it’s not.
Yeah he’s just demonstrating a sorting effect, calculating the magnitude would take a serious model (much easier to do it with code). That’s why I asked if he could do a 60% win rate mod, I wanted to see if the sorting was still obvious.
it takes hundreds, if not thousands, of parses to get a proper analysis.