Just bad luck?

[NineShadowEyes]

I've now run three sessions with FGII. Everything is going pretty smoothly. There's only two players, but they run 2 characters each. One of the players is having an unbelievable run of low rolls. The other player and myself get a wide range of numbers, but this guy, he consistently gets < 5. Not kidding, his roll history will look like this: 1, 1, 2, 3, 5, 6, 1, 3, 17, 1, 1, 3, 2, 2, 1, 1, etc... I don't really believe there's some bug doing it, but... what's up with that? It was funny for a while, but now it's to the point that I let him reroll every other 1 cause he rolls so many. He hasn't hit *anything* yet, with either character. I kept track of his rolls last night: 31 total d20 rolls. Only 7 were *above* 5. And 8 1s!!! WTF?!

:confused: :confused: :confused:

[Mellock]

I have that happening to me a lot. All I can say is that I think it's more of a stroke of bad luck than anything else. Now, in theory, there is no such thing as a "real" random number on a computer, but you also have to put in a bit of effort to make the algorithm (or function) end up being that biased. If he were to have it that bad, more people simply should have it that bad. I'm curious to see where it leads if you really count it out though. I usually think along the lines of people easily forgetting good rolls, and making a lot of effort to remember the bad ones. :)

Occasionally, I seem to make up for it by having a session where I can't do anything wrong though. That helps :D

This is also why I kind of like the "bennies" system, to at least have *some* counterweight for when things go *really* awry. Bunnies aren't supposed to cause a total party kill.

[Dachannien]

Back when our group was playing tabletop D&D, one of the guys had this evil evil d20 that rolled an unnatural number of 1's. Part of the problem was his strange die-rolling style, which resulted in the die doing kind of a yaw around the vertical axis rather than tumbling across the tabletop, but later one of the other guys took the die home and gave it a few hundred test rolls (much to his wife's chagrin, hehe :) ). Turned out that it rolled 1's and 19's each about twice as often as any of the other numbers.

What was even more amusing was that the guy who used this die was playing a character who had a flaming rapier. Ofttimes the rapier was the only light source the party was using, and so when he'd roll a 1, he'd have to make a fumble check (a dex check against a DC of 10). He'd end up rolling a 1 on that one as well, thus dropping his flaming rapier and plunging the party into darkness.

[Blue Haven]

I've now run three sessions with FGII. Everything is going pretty smoothly. There's only two players, but they run 2 characters each. One of the players is having an unbelievable run of low rolls. The other player and myself get a wide range of numbers, but this guy, he consistently gets < 5. Not kidding, his roll history will look like this: 1, 1, 2, 3, 5, 6, 1, 3, 17, 1, 1, 3, 2, 2, 1, 1, etc... I don't really believe there's some bug doing it, but... what's up with that? It was funny for a while, but now it's to the point that I let him reroll every other 1 cause he rolls so many. He hasn't hit *anything* yet, with either character. I kept track of his rolls last night: 31 total d20 rolls. Only 7 were *above* 5. And 8 1s!!! WTF?!

:confused: :confused: :confused:

I believe in you my friend...we had a game today and me and my friend, we were "cursed" by the FG dices, only low numbers as 1 and 2 and 4 :( this must be some bug on the generator or something :( i don´t understand how this works...and sometimes my friend cannot see the right number, like i roll a 6 and my friend see a 2...very strange...

[Hayt]

Well, it works both ways. The last session I played, nearly all my rolls aced - but I ain't complaining

[Hamish]

I rolled 30d20 10 times and the average was 10.36
Pretty normal I'd say.

[joshuha]

I am gonna write a script today to test rolling the dice 10,000 times and recording the average on a d20. I will use the throwDice command to replicate actually throwing the dice (as opposed to say using the lua math.rand).

I will post results later tonight.

[Mellock]

Oooh, let's hack out math.rand and replace it with a combined tausworthe generator! :D

(Isn't throwdice just a function that uses the lua math.rand?)

[joshuha]

That's the thing though, I am pretty sure that for its own dice FG doesn't use math.rand but some kinda of physics simulation. Thats why I want to use the throwDice function as it will replicate actually throwing the dice onto the chat window as opposed to just spitting out a number.

[Mellock]

Huh... that's odd. I have to say I *am* curious about it now.

[joshuha]

Using control.throwDice (which does it look like it throws it to random locations at the chatbox) I get the following totals:



Rolls Total Average
-------------------------
100 1081 10.81
1000 10259 10.259
10000 105096 10.5096


The average of a d20 roll should be around 10.5. So it does look like the die engine is decently random over time. Now whether or not their are abnormally long strings of bad/good rolls in there I have not analyzed (like say 15 1's in a row) but overall the engine doesn't seem to be weighted.

[Mellock]

Just out of curiousity: does that average on 10000 rolls mean that 5000 1's and 5000 20's could produce that result? :)

Would it be much work to script a counter that counted which dice are rolled how many times? (And it's just a question of out interest more than anything, I wouldn't want to inconvenience you by having you put more effort in it then you have already!)

[NineShadowEyes]

Yeah, get back to work and do a standard deviation script! Slacker! :D

Geez, I hope it doesn't get so bad that my pal rolls 15 1s in a row. I don't think his ego could take it.

Of course, the flip side to this is that one day he's going to roll nothing lower than 18 for several sessions.

BTW, we also get the glitch were the visible dice show a number, but it's actually different. Which is harsh cause once I saw him rolling a 20 and it turned out to be a 6.

BTW #2, just pointing out that the other player and myself enjoy a healthy spread of random numbers on the d20. It's just my pal who has the sucky d20 mojo. Which, of course, is ironic because he's always had mad d20 skillz with real dice.

[Sorontar]

Yeah, get back to work and do a standard deviation script! Slacker! :D

Geez, I hope it doesn't get so bad that my pal rolls 15 1s in a row. I don't think his ego could take it.

Of course, the flip side to this is that one day he's going to roll nothing lower than 18 for several sessions.

BTW, we also get the glitch were the visible dice show a number, but it's actually different. Which is harsh cause once I saw him rolling a 20 and it turned out to be a 6.

BTW #2, just pointing out that the other player and myself enjoy a healthy spread of random numbers on the d20. It's just my pal who has the sucky d20 mojo. Which, of course, is ironic because he's always had mad d20 skillz with real dice.

Just tell him to ditch the candyfloss pink dice :)

[joshuha]

Just out of curiousity: does that average on 10000 rolls mean that 5000 1's and 5000 20's could produce that result? :)

Would it be much work to script a counter that counted which dice are rolled how many times? (And it's just a question of out interest more than anything, I wouldn't want to inconvenience you by having you put more effort in it then you have already!)

In theory yes the average could still be a good average but the range could be skewed.

I ran the test 10,000 times again, this time recording number of times a number came up:

Rolls: 10,000
Total: 104791
Average: 10.4791

["dielist"] = {
[1] = 561,
[2] = 501,
[3] = 501,
[4] = 466,
[5] = 467,
[6] = 518,
[7] = 487,
[8] = 491,
[9] = 536,
[10] = 487,
[11] = 492,
[12] = 493,
[13] = 535,
[14] = 482,
[15] = 485,
[16] = 454,
[17] = 496,
[18] = 526,
[19] = 523,
[20] = 499,
}

[joshuha]

If you want to test this, in a d20 test ruleset I did the following:

In chatmanager.lua:



function test()
list = {};
for i=1, 20 do
list[i] = 0;
end

CampaignRegistry.dielist = list;
CampaignRegistry.Total = 0;
for i=1,10000 do
control.throwDice("dice", {"d20"}, 0, "Test");
end
end
-- Initialization
function onInit()
registerSlashHandler("/whisper", processWhisper);
registerSlashHandler("/die", processDie);
registerSlashHandler("/test", test);
end


In chat_chat.lua under the onDiceLanded function:


elseif draginfo.isType("dice") then
local dicetable = {};
dicetable = draginfo.getDieList();
list = CampaignRegistry.dielist;
list[dicetable[1].result] = list[dicetable[1].result] + 1;
CampaignRegistry.Total = CampaignRegistry.Total + dicetable[1].result;

--applyModifierStackToRoll(draginfo);
end


Note when doing this for 10,000 that FG will freeze up for a minute or two trying to graphical process all the rolls so just give it time.

*EDIT*

The above will essentially queue up all the dice to roll at once hence the freezing. I have also tested putting the throwDice on the onDiceLanded which does wait until one lands until it throws another but for large values this takes forever.

[Mellock]

Heh, so that's lua eh... it has a certain elegance to it. :)

And that pretty much proves the dice just aren't biased... Your friend should have some more beneficial statistics coming his way anytime soon, NineShadowEyes. :D

[richvalle]

Hey!!!! There is proof right there! The number 1 came up more then any other number.

FG cheats!

:)

rv

[Griogre]

Now, now no hasty conclusions! We need to study this issue carefully and fully - only more testing can reveal if the die is actually has a bias. ;) My study to do a study prices are quite reasonable. :) Oops, I apparently was party assimilated by the terminal inaction crowd and didn’t know it. :p

[Mellock]

Ooooh, or what if the dice *know* when they're being examined? Kind of like how Chucky turned into a regular doll when he was picked up!

[richvalle]

Hmmm... dice that are observed can't be measured?

Maybe that is why sometimes you can see a 6 on the screen but a 1 in the logs.

Crafty dice!

rv

[joshuha]

One more test, larger sample size:

["Total"] = 1.05066e+006,
["numrolls"] = 100000,
["dielist"] = {
[1] = 5148,
[2] = 5216,
[3] = 4845,
[4] = 4888,
[5] = 5136,
[6] = 5006,
[7] = 4912,
[8] = 5060,
[9] = 5004,
[10] = 4849,
[11] = 4658,
[12] = 4963,
[13] = 5040,
[14] = 5074,
[15] = 4928,
[16] = 4902,
[17] = 4943,
[18] = 5125,
[19] = 5155,
[20] = 5148,
}

[Blue Haven]

Just tell him to ditch the candyfloss pink dice :)

I already tryed that :( no result...

[Mellock]

["numrolls"] = 100000,
[1] = 5148,
[20] = 5148,


:eek:

Hahaha, 100.000... that *is* thorought though! :D

[Dachannien]

One thing I would point out is that the throwDice function doesn't exactly simulate the same process as the user picking up a die and rolling it (especially if the user picks up the die and just sort of drops it on the chat window*).

It's not clear to me whether the randomization of the die that occurs when you grab hold of it with the mouse cursor is the same as the randomization that occurs on the die when you use throwDice to let the computer do the rolling. What I mean is, when you pick up a die manually with the mouse, is it truly randomly presenting an orientation of the die to you, as presumably happens when the computer does the rolling?

If the same process is used in both cases, then Joshuha's experiment seems conclusive enough to me. If it's not, there could be some explanation for strings of bad rolls related to the specific way that some players are rolling the dice.

The SmiteWorks guys could probably offer some clarification on this, of course.

(* I'm reminded of the Super Bowl a couple years ago when the honorary ten-year-old coin tosser pancaked the coin toss, earning the sarcastic derision of one of the players standing next to him: "Nice toss, kid.")

[richvalle]

From what I've seen, the die is randomized as to what face is up when you first grab the die. At least that is what I think it is doing.

rv

[Griogre]

I actually played around with this a while back. Probably rolling the dice can skew the results if you just drop it. It would be interesting to know if the original poster's guy with horrible rolls was using the mouse to roll or the finger panel on a laptop. If he was I would tell him to make /die codes instead of rolling his dice. :)

[Callum]

One more test, larger sample size:

["Total"] = 1.05066e+006,
["numrolls"] = 100000,
["dielist"] = {
[1] = 5148,
[2] = 5216,
[3] = 4845,
[4] = 4888,
[5] = 5136,
[6] = 5006,
[7] = 4912,
[8] = 5060,
[9] = 5004,
[10] = 4849,
[11] = 4658,
[12] = 4963,
[13] = 5040,
[14] = 5074,
[15] = 4928,
[16] = 4902,
[17] = 4943,
[18] = 5125,
[19] = 5155,
[20] = 5148,
}
Hmm...those numbers give a chi-squared value of 72.56, which equates to a probability of less than 0.01% that those rolls would occur if the die was truly random.

Maybe some genuine statisticians would like to correct me, but that doesn't look good!

On the other hand, the chi-squared value for joshuha's earlier set of 10,000 rolls is only 26.46 - a probability of about 12% that such a set of rolls would occur. I'd like to see some more tests...

[richvalle]

You lost me on that one Callum.

rv

[Mellock]

Hmm...those numbers give a chi-squared value of 72.56, which equates to a probability of less than 0.01% that those rolls would occur if the die was truly random.

Maybe some genuine statisticians would like to correct me, but that doesn't look good!

Heh, the best you usually do with a computer is pseudo-random really. You can get it more "real" random by incorporating things like athmospheric noise and stuff, but in the end, I think in this case the result is more than satisfactory.

Or perhaps the dice really *do* know when they're being watched! Oooh, spooky! :ogre:


You lost me on that one Callum.

rv

Imagine that every roll had come up exactly 5000 times. On one hand, that would have been a "perfect distribution". You also would have immediately thought that there was no way that those results could have been "really" random. One possible way of coming to that distribution would have been if you just rolled the same sequence over and over again 1, 2, 3,..., 20. for instance.

[Callum]

You lost me on that one Callum.
Sorry about that! The basic idea is this: if you rolled a d20 100000 times, you'd expect each number to come up 5000 times. As you can see from joshuha's list, that doesn't happen - there's quite a lot of deviation from that ideal distribution (including the shocking result for 11, with only 4658 appearances). The chi-squared value is a measure of how great that deviation is. We can then calculate the chance of seeing that level of deviation if the events were truly random. A 0.01% probability is very statistically significant, while a 12% probability isn't.

[Hamish]

I don't have the time to test it right now, but I seriously doubt any physical die is going to achieve better results. But I have to admit, I swear by lucky dice too. :)

[sloejack]

Sorry about that! The basic idea is this: if you rolled a d20 100000 times, you'd expect each number to come up 5000 times. As you can see from joshuha's list, that doesn't happen - there's quite a lot of deviation from that ideal distribution (including the shocking result for 11, with only 4658 appearances). The chi-squared value is a measure of how great that deviation is. We can then calculate the chance of seeing that level of deviation if the events were truly random. A 0.01% probability is very statistically significant, while a 12% probability isn't.

Dude, I think you've gone off into the weeds on this one. The odds of rolling any single number on a 20 sided die is 5% assuming that the die is rolled fairly and in the same manner each time. Looking at the successive test results that were posted you see that as more rolls were executed the closer the frequency got to expectation. With a deviation at 100000 rolls of 0.6% which is improved over the deviation of 1.1% over 10000. I wondered myself where the intersection would be of # of rolls required for each number to come up the exact same amount of times. It's certainly north of one million, though I suspect that unless you watched and validated the results of each individual roll, you may never find that intersection and frankly it could happen much sooner. I think that past 1000000 rolls you would certainly see the frequency of appearance to be even closer to the expected 5%.

I think we can all agree that over a significant amount of time and rolls, all of the numbers will come up at about the same frequency. The question of is the die roller fair in this case I believe the answer to be yes.

Why people generally attribute clusters of results to curses, luck, streaks etc, is that the gaming session rarely has more than lets say 1000 rolls of a single die type, and no one keeps track of all of the results of every die roll they've ever made or observed since the time they started rolling dice to see that the law of large numbers holds true which I think is much more relevant to the question at hand.

[DNH]

Aha! Finally, someone pointed out that the probability of any given number coming up on a d20 is exactly 1 in 20, or 5%. And while we're at it, there is no average on a 1d20 roll. Not 10.5 or 10, or 11. Each number is equally likely to occur. Take two d20 though, and that's a different story.

My take on all this is that it is down to perception. It's like when you're walking down a busy street and you're in something of a hurry and you marvel at how everyone, and I mean EVERYONE, is coming in the opposite direction and getting in your way. It's not true; it just looks that way. Turn around and go back the other way and you will see the exact same thing happening!

[joshuha]

More data:

["Total"] = 1.05017e+007,
["numrolls"] = 1e+006,
["dielist"] = {
[1] = 51367,
[2] = 52417,
[3] = 49023,
[4] = 48252,
[5] = 51327,
[6] = 49664,
[7] = 49401,
[8] = 49839,
[9] = 50062,
[10] = 48762,
[11] = 46930,
[12] = 49640,
[13] = 51267,
[14] = 51369,
[15] = 49191,
[16] = 48734,
[17] = 48805,
[18] = 51375,
[19] = 51119,
[20] = 51456,
},

[sloejack]

I ran the same test for 1 mil rolls and ended up fairly close to your results. The interesting anomaly I noticed was that 11s for you and for me were the lowest in frequency, 94280 (4.71%) times, and 2s coming up the most frequently, 104425 (5.22%) times, out of 2 million rolls.

For what it's worth here are the total findings:


Number Rolled Percent
2 104425 5.22
20 103102 5.16
5 102989 5.15
1 102637 5.13
14 102595 5.13
18 102585 5.13
19 102093 5.10
13 102036 5.10
9 100077 5.00
8 99714 4.99
6 99462 4.97
12 99311 4.97
15 98875 4.94
7 98606 4.93
16 97997 4.90
3 97693 4.88
10 97539 4.88
17 97433 4.87
4 96551 4.83
11 94280 4.71
2000000 100

[Kalan]

Gah...must be Christmas season :D All these coding guru's goofin around while they supposed to be working :D

God Bless America ;)

[Dachannien]

Aha! Finally, someone pointed out that the probability of any given number coming up on a d20 is exactly 1 in 20, or 5%. And while we're at it, there is no average on a 1d20 roll. Not 10.5 or 10, or 11. Each number is equally likely to occur. Take two d20 though, and that's a different story.

I think when people say that the average on a d20 is 10.5, they mean that the expected value of a die roll on a fair d20 is 10.5, which is true.

[NineShadowEyes]

I actually played around with this a while back. Probably rolling the dice can skew the results if you just drop it. It would be interesting to know if the original poster's guy with horrible rolls was using the mouse to roll or the finger panel on a laptop. If he was I would tell him to make /die codes instead of rolling his dice. :)

He's using hotkeys.

[Griogre]

I'd have to say he's unlucky then. ;) Does he use the /die command or drag and drop the to hit die from the weapons sheet to the hotkey? To make him feel better tell him to do it the other way. :p

[Blue Haven]

And how about the problem that shows one number on the dice and another on the chat window??? like we had a game today and that crap happens a lot of times, but the person that throw the dice cannot see the real roll only the others see the real number...

[richvalle]

It happens in our game from time to time. Maybe about 1 time every 3 or 4 games.

I have a standing rule that the number that shows up in the chat window is the 'real' number.

rv

[Blue Haven]

Humm...so if you and more ppl see this problem...why the hell Devs don&#180;t fix it??
Because selling a product that have problems is kind of...you tell me if i sell you something that works bad...and you will know what i mean...
This crap never happen on FG 1.05f so why this is happening now?? lol always the same...buy it good good...we have a problem, ahh do this maybe it work i don&#180;t care... :( it&#180;s a shame...

[richvalle]

Hmmm, I'm not sure it didn't happen in FG1.

I guess we have a differenace of opinion here. To me its not a MAJOR "Oh my god the game is broken!!!" kind of problem. Its a minor annoyance. Maybe the devs are actually trying to enjoy Christmas. :)

rv

[Blue Haven]

Hmmm, I'm not sure it didn't happen in FG1.

I guess we have a differenace of opinion here. To me its not a MAJOR "Oh my god the game is broken!!!" kind of problem. Its a minor annoyance. Maybe the devs are actually trying to enjoy Christmas. :)

rv

i didn´t say that is a Major Problem...but like i said, you pay to use this program and like anything else you pay to use, you don´t like when it has problems that annoy the flow of the game...
Merry Christmas to you Richvalle and i hope they fix this crap :) :rv:

P.S.- this never happen in FG 1.05f :)

[Sorontar]

Yeah it's certainly not a show stopper, just go with the chat window.

I am sure that the devs know about it but it is down the list of things to fix, personally I'd rather them fix the Save Campaign issue first so we don't lose data on crashes even if there is a workaround.

There are bigger fish to fry than a graphical glitch IMHO.

I have complete faith that the guys at Smiteworks are working towards fixing the issues and improving the product further.

Hope they have a great Christmas and that Santa brings them something special :)

[Dachannien]

I'm pretty sure I've seen the dice and the chat window result come up with different numbers in 1.05. Just use richvalle's house rule - problem solved. Regardless, that's a minor issue compared to other bugs with the software.

[joshuha]

It did occur in 1.05 and its due to a synching issue. I think I know some reasons why it may occur but I need to do some more testing.

[Griogre]

Like richValle our standing rule is the number rolled is the one that shows up in chat. This did happen in 1.05f and I believe the developers mentioned the problem. It seems to happen most when things are laggy.

[Tokuriku]

Just a tought...

Maybe the seeding for the random function happens when you start FGII and each time it's different. So your tests would be pointless unless you happen to have, like that guy, the same seed. It could also be a problem of cpu speed dropping the ball when the random number must be decided. What I meanis, the person with the problem must do the tests him/herself. Then we'd see if there really is a problem. But I'm inclined to believe that it is the host computer that decides the randomness of a die so there again, we don't know for sure what to measure.

Maybe we should ask the devs for some insight and then be sure to test the right thing...

[Griogre]

The seed would be on the master not on the slave. I believe the tests were valid because they were testing if the pseudo random number generator was "random enough" not whether every seed had about the same randomness. We know by definition on a pseudo random generator they won't. Other than the 2's the data looked pretty random to me.

Conceptually the player is getting killed on runs. That is not really true either because he is not rolling all his “to hits” at the same time – other players and the DM are rolling in between his rolls. Now as a DM I have seen strings of bad rolls – runs – where everyone rolls low for a round but not for what I would think would be a statistically long period of time. The same for high rolls too.

I think it is human nature to remember the very high and low rolls and forget the middles where most of the dice fall. Thus if you roll a 1, 10, 4, 12 people tend to remember the 1 and 4. I also think it is human nature that once a person has a “rep” for a high or low roller when that type of roll happens people remember the rolls that match the rep more because that is what they expect, so they just validate their expectation. That’s why I believe the statistics over any human perception in this particular case.

[NineShadowEyes]

The bad mojo has left the building, apparently. The player in question rolled a fairly wide spread of numbers during the last session.

Still seeing the dice show one number and the chat window show another, though.

[richvalle]

i didn´t say that is a Major Problem...but like i said, you pay to use this program and like anything else you pay to use, you don´t like when it has problems that annoy the flow of the game...
Merry Christmas to you Richvalle and i hope they fix this crap :) :rv:

P.S.- this never happen in FG 1.05f :)


Hey! Merry Xmas to you too! Well, I hope it was a good one since it is now after the fact. :)

rv

[Blue Haven]

Hey! Merry Xmas to you too! Well, I hope it was a good one since it is now after the fact. :)

rv

Ok Thanks :) Merry Christmas to you and a Happy New Year :D without crappy dice!! lol

[NineShadowEyes]

Just a quick note to say the guy who was rolling all the 1s rolled four 20s last session.