|
Post by greenjello on Jan 14, 2018 16:25:31 GMT
In the first test i would suspect that the die was broken and not rolling enough 1s. If i had only preformed the fifth test though, i could suspect it was MORE likely to roll a 1 than it was supposed to. In the first test i got 33% fewer 1s than expected. In test 5 i got 24% more than expected. Now this is the SAME die, so if it is biased, it is as biased in the first test as in the fifth, right? If i had only rolled 30 times rather than 200 i would have gotten even more varying results. Also, in the 4th test the die was 30% more likely to roll a 4 than expected. In the 5th test it was 24% LESS likely to roll a 4 than expected. If I'm understanding your chart correctly, I'd be fine with the results from 200 rolls. All of your examples are with 5% of the expected value of 16.6% of the rolls. Really for me it comes down to good enough. 30% off is pretty bad, and too much of a variation to be tolerated, but 5% is fine. I might toss out some dice that are fine, but dice are cheap, while my time is not. As for the age of this die, i have no idea. It is super unlikely that is it from the same batch of dice used in the article, but it has the same rounded edges (allegedly terrible for randomness) and it even has some visible flaws on the surface. Could be there were more issues than just the rounded edges with the dice in the article. You'll notice that even after "fixing" the edges, they're still not perfect. What leads you to this assertion? Adjusting the weight of dice is a time honored way of loading them. Generally the weight is a bit more dramatically shifted, so it probably depends on the size of the void how much it effects things. FWIW, this was originally posted on Warseer, there's link to it in the Dakka Dakka article. www.warseer.com/forums/showthread.php?65531-Do-you-roll-a-lot-of-1s-A-40k-must-readThe original also doesn't seem to have any links to sponsors. FWIW, I suspect that most people would just buy their dice from whatever turned up first in a google search. Anyway, there's more discussion on there if you're interested. Further down in the replies to the original Yade (the author) mentions: So really the rounding appears to be all about the Benjamins! Also in the thread is a link to another study, where two opponents wrote down the results of 6 40K games, and came away with recorded proof that the "lucky" player, was in fact 15% more likely to have favorable results. www.warseer.com/forums/showthread.php?139400-Is-my-opponent-just-luckier-than-meFinally a short method for quickly checking to see if your dice are biased: www.warseer.com/forums/showthread.php?65531-Do-you-roll-a-lot-of-1s-A-40k-must-read&p=1302000&viewfull=1#post1302000Just another observation: with Leadership tests removed (good riddance) there is now no reason to roll low in Warmachine. One of the best ways to force people to deal with bias in their dice is to make them use the same set throughout a game, and not swap out. Not sure where I'd suggest re-introducing a low roll wins mechanism in the game.
|
|
|
Post by deviantcad on Jan 14, 2018 16:26:15 GMT
The article is interesting, it starts with 3 links promoting one dice, but they end up choosing one with an obvious flaw to test. They do skip over the fact that the promoted dice was hard to read and say in spite of any evidence the that it is probably better or would be if they didn't pick one with an obvious flaw.
Again the RNG in exel is fine for these purposes. Unlike a game designer it doesn't have an incentive to skew the numbers. I don't have trollock's gumption, but if you want to write a RNG program go for it.
I have seen precision milled dice that are pretty cool.
Just saw the 2 articles you added. The first one doesn't give any supporting evidence and the second suggest a high number of rolls then the initial article you posted.
|
|
|
Post by Trollock on Jan 14, 2018 21:00:21 GMT
I did some more Excel work lol. I coded a d6 that was NOT biased and ran the 200 rolls-test on it, exactly as i did with my physical die. After the full 1000 rolls the results were slightly better than the physical die. The chance of each face coming up varied from 15.6% to 17.3%. With the physical die it varied from 15.6% to 18.1% so the span was slightly larger, but not extremely so. What was interesting is that if i check after 200 rolls five times with the excel simulation, the results look at least as biased as i got with my physical die. 200 rolls does not seem to be enough with a "truly random" die. 1000 rolls seem more reliable. greenjelloIm not really worried about your wallet man. If you want to throw away dice that are perfectly fine, i wont stop you. Im just questioning if the method you use is valid. If it can cause you to throw away dice that are actually fine, it can cause you to keep dice that ARE biased as well, since it is so hard to detect bias. I am seriously thinking about whipping out MATLAB and start coding dice that are biased and see if i can detect the bias with a test, but this is staring to become way more time consuming than is probably healthy
|
|
|
Post by greenjello on Jan 14, 2018 22:59:00 GMT
And what do you base that assertion on? Excel has no incentive to skew the numbers, but they're not necessarily producing the sort of random that people expect from dice. Unnatural runs are one possible problem, there are others. Trollock doing more rolls with avoid some problems with some algorithms which need to "warm up", and also tend to produce unnatural runs of numbers. But don't just take my word for it, here's another paper. www.pages.drexel.edu/~bdm25/excel2007.pdfEDIT: Better paper on the problems in Excel: blog.richpollock.com/2014/08/randomness-in-excel/Here's some relevant information about flaws that have appeared in Excel. Just to note, these are programs written by professionals, supposedly checked by QA, and then released to the millions of users. And still flawed. Do you really think your average Joe is going to get this right most of the time? Supposedly this has been fixed in Excel 2010, so maybe your results are fine. I think the real take-away here is that random on the computer is a hard problem. I think you and I are disagree on how many rolls are necessary to get a decent result. You're correct of course that the more die rolls, the more accurate the results. However, I'm fine with a +-5% skew, which shows up at 200 rolls, and it appears is possible with even fewer. Much more than that, and all the other problems I have a probably going to swamp it.
|
|
|
Post by greenjello on Jan 14, 2018 23:06:27 GMT
|
|
|
Post by Trollock on Jan 15, 2018 7:18:20 GMT
I use the latest version of excel just for the record.
|
|
|
Post by Trollock on Jan 15, 2018 7:39:30 GMT
And here is the same analysis preformed on random numbers generated by www.random.org/200 rolls gives wild fluctuations. 30 rolls is insanity tbh.
|
|
|
Post by greenjello on Jan 15, 2018 14:59:35 GMT
Hmm, looks like you pic isn't show up. What did you put into the generator? I did a bit of poking about with it, and it appears you're correct, 30 is too low. It's not going to settle down to a nice distribution until around 100-200 rolls. At 90s rolls the distributions are within 5% of average. At 210 rolls they're within 1-2%. drive.google.com/open?id=1UcnTJZiqsUVczy3WJ-PCEnfuHY04fWsIUYoBPoZ_OPoAlso not sure if you're getting the point about Excel, and psuedo-random number generators in general. It's not a certainty that the latest Excel doesn't have yet another bug introduced into it. Really only Excel2010 appears to have been certified to be working correctly, anything else is suspect.
|
|
|
Post by Trollock on Jan 15, 2018 15:08:22 GMT
Hmm, looks like you pic isn't show up. What did you put into the generator? I did a bit of poking about with it, and it appears you're correct, 30 is too low. It's not going to settle down to a nice distribution until around 100-200 rolls. At 90s rolls the distributions are within 5% of average. At 210 rolls they're within 1-2%. drive.google.com/open?id=1UcnTJZiqsUVczy3WJ-PCEnfuHY04fWsIUYoBPoZ_OPoAlso not sure if you're getting the point about Excel, and psuedo-random number generators in general. It's not a certainty that the latest Excel doesn't have yet another bug introduced into it. Really only Excel2010 appears to have been certified to be working correctly, anything else is suspect. Damned photobucket requiring me to pay money... it is fixed now. I get your point about excel. I do not think it matters here, but i get your point. random.org uses atmospheric disturbances to generate their random numbers, so they are "truly random". The results of both the excel simulation and random.org are basically the same. At 200 rolls there are still cases where you get wild fluctuations. Some of them were worse than what i had with my physical die.
|
|
ravenfire
Junior Strategist
Me Mulg me SMASSSSSHHHH
Posts: 244
|
Post by ravenfire on Jan 15, 2018 15:16:59 GMT
I think you need to take in to account the standard deviation of a side to determine if the result was probable. It might be that 28 rolls of a 1 in Test 1 seems low, but still is within the tolarance of your standard deviation I hope I am making some sense.
|
|
|
Post by Trollock on Jan 15, 2018 15:17:10 GMT
greenjelloOh, i think you and i think of different things when you say the distribution is within 5%. Do i understand you correct that you mean that the distribution is within 5% if the chance of each face showing up is somewhere between 11.7-21.7% (16.7% ± 5%)? I interpreted "within 5%" to be more like 15.83-17,5%. Just so i know what we are talking about
|
|
ravenfire
Junior Strategist
Me Mulg me SMASSSSSHHHH
Posts: 244
|
Post by ravenfire on Jan 15, 2018 15:18:08 GMT
to be sure: what I mean is that X is the perfect number of 1's you should roll, but you may be X-5% and X+5% and still have a valid test result.
But this only says something about the probability that your testresult is the way it is.
You can still roll 1000 1's in one streak, and the die may still be perfect. It is just that this is very unlikely to happen, but it can happen
|
|
|
Post by greenjello on Jan 15, 2018 15:29:01 GMT
Yeah, saw Random.org a few times while looking other stuff for this. I like it better than any computer generated numbers, very cool idea for getting something random in a way that's distributed, and produces a ton of results.
I also must admit that after having read a number of articles like the ones posted I also tend to be very biased against any computer RNG, so Excel might be fine, but I don't trust it.
I've been trying to say the same thing as Ravenfire, X+5% or X-5%. Getting a reasonably fair dice roll is within +/-5% of the 16.6% distribution dictated by a perfect random source.
After that much I really do feel like the dice rolls can easily be swamped out by other personal failings. Order of activation, brain farts, poor list selection, or poor decision making. One thing I've always liked about the Warmachine system is the probability mitigation rules: Boosting, puppet strings, Calandra's feat, etc. Tips the skill/luck balance much more firmly into the skill column.
EDIT: Skill not luck column, makes a difference.
|
|
ravenfire
Junior Strategist
Me Mulg me SMASSSSSHHHH
Posts: 244
|
Post by ravenfire on Jan 15, 2018 16:41:26 GMT
a positive attitude.. yes you can influence the dice
|
|
|
Post by elwinar on Jan 16, 2018 9:30:50 GMT
From a philosophical point of view, is there really any way of producing true random ? A computer certainly cannot, and (surprise) your perfect dice cannot either as you could guess the side it will fall on based on how you throw it, distance to the table, etc. The point being that randomness as we see it in day to day life is much more about unpredictability than probability. Given that logical affirmation, and given the fact that we roll a low number of dice in a game (statistically anyway), if you cannot identify a dice as crooked in the span of a game it probably doesn't matter if it is or not. At this scale, you can't really differentiate "luck" from bias, so the main condition (unpredictability) is fulfilled.
The good news is that Warmachine is much more of a skill game than a chance game, so biased dice aren't really a problem. Also, I've read the post DeviantCad referenced earlier, and I agree with the argument it made: cheating is a matter of intention rather than result. If you knowingly use crooked dice, you're cheating. If you do voodoo magic to have better rolls, whether or not it worked is irrelevant because it's cheating nonetheless.
That being said, I prefer and use sharp-edged casino dice. The reason I prefer them is because they tend to get out of the table less often, are easier to read from afar so your opponent never have to take your work on the result of a dice and you can find them on the floor quickly when you drop them.
|
|