Tom -
As a nice exercise, maybe you can automate a routine that checks the dice rolls in every game log you have. I have a strange feeling that the game rolls a larger number of 1's than expected
-Chet
Random Number Generator
Moderator: TFF Mods
-
- Rulz Guru
- Posts: 801
- Joined: Thu Jan 01, 1970 12:00 am
- Location: Amherst, NY
- Contact:
Random Number Generator
Reason: ''
- GalakStarscraper
- Godfather of Blood Bowl
- Posts: 15882
- Joined: Tue Jun 26, 2001 12:00 am
- Location: Indiana, USA
- Contact:
Pink Horror and I are working on generating a sample to statistically test the rolls to valid the game with short term dice analysis.
It's going to take a while but we are in the process of double checking the randomizer.
We've tested the randomizers both for D6 and 2D6 rolls over a 3 sets of 100,000 and it came out fine statistically on all 6 samples.
Pink and I and now running sets of 30 rolls to test the probablity of the 20 groups of sequential rolls of 6 dice in this set (see PinkHorror's know buried post on probabilities of chains of 6 rolls of dice on the BBC ( http://pub10.ezboard.com/fbloodbowlcent ... ic&index=3 )). We will probably run 200 to 400 sets to test the the groups in the set and then the probablilites in the only the first 6 of each set as well.
Its in motion Chet, its just time consuming to reload the program 400 times to generate all the samples. We confirmed that the randomizer is working fine long term. We are now verifying that it works fine short term. Also keep in mind that the same routine that rolls the D6, rolls the Block dice and I'm pretty sure that 6 is linked to Pow and I've seen a lot of complaints about those as well. At the end of the day, it may come down to human perception.
Galak
It's going to take a while but we are in the process of double checking the randomizer.
We've tested the randomizers both for D6 and 2D6 rolls over a 3 sets of 100,000 and it came out fine statistically on all 6 samples.
Pink and I and now running sets of 30 rolls to test the probablity of the 20 groups of sequential rolls of 6 dice in this set (see PinkHorror's know buried post on probabilities of chains of 6 rolls of dice on the BBC ( http://pub10.ezboard.com/fbloodbowlcent ... ic&index=3 )). We will probably run 200 to 400 sets to test the the groups in the set and then the probablilites in the only the first 6 of each set as well.
Its in motion Chet, its just time consuming to reload the program 400 times to generate all the samples. We confirmed that the randomizer is working fine long term. We are now verifying that it works fine short term. Also keep in mind that the same routine that rolls the D6, rolls the Block dice and I'm pretty sure that 6 is linked to Pow and I've seen a lot of complaints about those as well. At the end of the day, it may come down to human perception.
Galak
Reason: ''
-
- Super Star
- Posts: 1042
- Joined: Thu Jan 01, 1970 12:00 am
- Location: Surrey
PBEM Dice
Acerak,
Back at the end of last year, similar queries were raised which prompted Galak to roll 100,000 dice 3 times - he assured us he didn't lose count!
> Okay, I ran each one of the three methods through
> 100,000 dice rolls 3 times for each method for each
> of the D6 and 2D6 rolls ..... I'll let someone with
> better statistics figure out if it all means anything.
>
> Father Tim, Martyn, or anyone else you wants
> to look it over here you go.
Father Tim ran some tests and replied:-
Okay, I crunched the numbers and ran a few tests, including chi-squared. At the 0.001 level of significance (meaning 999 times out of 1000) all three algorithms produce near-perfect distibutions.
Erwin raised an excellent point, though. We need to look at the order in which each result occurs with each algorithm. Unfortunately, I can't think of any way to do this other than sequentially examine 100,000 rolls.
So in terms of total rolls all 3 options seem to give roughly the right distribution of results. However, as far as I'm aware, no test was done on the sequence of the numbers to see if there was any problem there. Eg if the program always rolled the following sequence "1,2,3,4,5,6,1,2......", statistically the results would be accurate in terms of each roll - however, the rolls could never be said to be random - which obviously is crucial for a random number generator.
FYI, the number of ones rolled in total for all algorithms combined - was 150,141 from 900,000 dice rolls - 16 more than expected - presumably these 16 are filtering into your games! Of the 3 methods, Brian's algorithm rolled the lowest number of 1s (49,879 from 300,000 rolls) - 121 fewer than expected - perhaps you should change to that algorithm! (although Tom's method rolled 81 less than expected!). Incidently both Brian's and Tom's algorithms rolled more than the expected number of 6s...... the standard Delphi rolled more 1s and less 6s - perhaps you are using this method!!!
Here is the summary of the rolls (sorry they don't tab very well):-
Dave
(Galak edited Dave's post to help line up the numbers for him)
Back at the end of last year, similar queries were raised which prompted Galak to roll 100,000 dice 3 times - he assured us he didn't lose count!
> Okay, I ran each one of the three methods through
> 100,000 dice rolls 3 times for each method for each
> of the D6 and 2D6 rolls ..... I'll let someone with
> better statistics figure out if it all means anything.
>
> Father Tim, Martyn, or anyone else you wants
> to look it over here you go.
Father Tim ran some tests and replied:-
Okay, I crunched the numbers and ran a few tests, including chi-squared. At the 0.001 level of significance (meaning 999 times out of 1000) all three algorithms produce near-perfect distibutions.
Erwin raised an excellent point, though. We need to look at the order in which each result occurs with each algorithm. Unfortunately, I can't think of any way to do this other than sequentially examine 100,000 rolls.
So in terms of total rolls all 3 options seem to give roughly the right distribution of results. However, as far as I'm aware, no test was done on the sequence of the numbers to see if there was any problem there. Eg if the program always rolled the following sequence "1,2,3,4,5,6,1,2......", statistically the results would be accurate in terms of each roll - however, the rolls could never be said to be random - which obviously is crucial for a random number generator.
FYI, the number of ones rolled in total for all algorithms combined - was 150,141 from 900,000 dice rolls - 16 more than expected - presumably these 16 are filtering into your games! Of the 3 methods, Brian's algorithm rolled the lowest number of 1s (49,879 from 300,000 rolls) - 121 fewer than expected - perhaps you should change to that algorithm! (although Tom's method rolled 81 less than expected!). Incidently both Brian's and Tom's algorithms rolled more than the expected number of 6s...... the standard Delphi rolled more 1s and less 6s - perhaps you are using this method!!!
Here is the summary of the rolls (sorry they don't tab very well):-
Code: Select all
Delphi Brian Tom
Set 1st 2nd 3rd 1st 2nd 3rd 1st 2nd 3rd
1 16,711 16,775 16,855 16,337 16,784 16,758 16,669 16,664 16,588
2 16,597 16,949 16,731 16,695 16,898 16,571 16,362 16,841 16,893
3 16,616 16,544 16,622 16,721 16,637 16,693 16,950 16,563 16,666
4 16,772 16,736 16,595 16,882 16,594 16,599 16,524 16,575 16,657
5 16,646 16,363 16,721 16,696 16,560 16,523 16,651 16,561 16,606
6 16,658 16,633 16,476 16,669 16,527 16,856 16,844 16,796 16,590
100,000 100,000 100,000 100,000 100,000 100,000 100,000 100,000 100,000
Dave
(Galak edited Dave's post to help line up the numbers for him)
Reason: ''
-
- Rulz Guru
- Posts: 801
- Joined: Thu Jan 01, 1970 12:00 am
- Location: Amherst, NY
- Contact:
- Trambi
- Ex-Mega Star, now just a Super Star
- Posts: 1310
- Joined: Thu Jan 01, 1970 12:00 am
- Location: St Quentin en Yvelines near Paris, France
- Contact:
-
- Rulz Guru
- Posts: 801
- Joined: Thu Jan 01, 1970 12:00 am
- Location: Amherst, NY
- Contact: