rollup pool question(s)

Posted: Mon Aug 22, 2011 4:40 pm

It would be neat to see if any of the Tourney managers who DA a lot, could add their data to that above. The larger the set of data, the more accurate the analysis.

I beleive that OAB's original analysis is most likely accurate. The averages are too close to each other to be anything else. I may actually look to maintain my spreadsheet more accurately so I can continue to check if this is true.

I also wonder how the Team Setup pool is? I wonder if the same averages would play out there. More managers should have accurate data for this to be done. Anyone else care to share? I can throw in about 40 setups or 200 warriors.

Posted: Mon Aug 22, 2011 6:54 pm

One could also go to the RU section on the forumns and get a lot of data just by going back and back and back Shocked

that's a lot of RUs

Posted: Mon Aug 22, 2011 7:53 pm

gentleben wrote:

One could also go to the RU section on the forumns and get a lot of data just by going back and back and back Shocked

that's a lot of RUs

true but there is probably some bias as not many post their truly horrible...

Posted: Tue Aug 23, 2011 1:34 am

Madwand wrote:

It would be neat to see if any of the Tourney managers who DA a lot, could add their data to that above. The larger the set of data, the more accurate the analysis.

I beleive that OAB's original analysis is most likely accurate. The averages are too close to each other to be anything else. I may actually look to maintain my spreadsheet more accurately so I can continue to check if this is true.

I also wonder how the Team Setup pool is? I wonder if the same averages would play out there. More managers should have accurate data for this to be done. Anyone else care to share? I can throw in about 40 setups or 200 warriors.

We just want to clarify that, despite the premise above (see bolded), we hardly, if ever, DA.
We also would not really call ourselves "tourney managers" as we are so actively devoted to the arenas.
Yet we did submit a lot of warrior data.

Posted: Tue Aug 23, 2011 1:38 am

Dark_Angel wrote:

gentleben wrote:

One could also go to the RU section on the forumns and get a lot of data just by going back and back and back Shocked

that's a lot of RUs

true but there is probably some bias as not many post their truly horrible...

The 100+ we posted were ALL that we received - the good, the bad, and the ugly.

Posted: Tue Aug 23, 2011 2:57 am

Dark_Angel wrote:

gentleben wrote:

One could also go to the RU section on the forumns and get a lot of data just by going back and back and back Shocked

that's a lot of RUs

true but there is probably some bias as not many post their truly horrible...

Not many post their truly great RUs either but it is still data. Lots of team sheets as well.

Posted: Tue Aug 23, 2011 6:39 am

The Consortium wrote:

Madwand wrote:

It would be neat to see if any of the Tourney managers who DA a lot, could add their data to that above. The larger the set of data, the more accurate the analysis.

I beleive that OAB's original analysis is most likely accurate. The averages are too close to each other to be anything else. I may actually look to maintain my spreadsheet more accurately so I can continue to check if this is true.

I also wonder how the Team Setup pool is? I wonder if the same averages would play out there. More managers should have accurate data for this to be done. Anyone else care to share? I can throw in about 40 setups or 200 warriors.

We just want to clarify that, despite the premise above (see bolded), we hardly, if ever, DA.
We also would not really call ourselves "tourney managers" as we are so actively devoted to the arenas.
Yet we did submit a lot of warrior data.

I never thought you were. I was hoping someone like Grimm may want to share his data, if he even keeps it. Or another who DA's a lot and is likely to have a large pool of data to add to what you and I posted earlier. My 30+ I posted are from the last couple of years but I have been lax in maintaining my database over the last 6-9 months.

This discussion is renewing my commitment to record ALL of my replacements and I will continue to measure the averages as OAB has done. I find it very interesting to see how close these numbers are to each other with the exception of size.

Posted: Tue Aug 23, 2011 6:47 am

The Consortium wrote:

Dark_Angel wrote:

gentleben wrote:

One could also go to the RU section on the forumns and get a lot of data just by going back and back and back Shocked

that's a lot of RUs

true but there is probably some bias as not many post their truly horrible...

The 100+ we posted were ALL that we received - the good, the bad, and the ugly.

Like Consortium, mine included ALL replacements I have gotten since I came back to the game with the exception of the last 6-9 months.

Posted: Tue Aug 23, 2011 12:21 pm

I delved into this issue a few years ago.
After data collection and reverse-engineering, there was one model for a rollup generator which matched the available data best.

Step 1: roll for size, normally distributed with a mean of 12 and a standard deviation of 4.
Step 2: randomly allocate 1 point at a time to the remaining stats until there are 70 points total.

I do not know what method the actual rollup generator uses for exception handling when a stat ends up with a value less than 3 or more than 21.
These happened about 1% of the time using this algorithm.

This "snowflake" method was probably used due to the poor PRNGs (pseudo-random number generators) which existed during the early days of Duelmasters.

Here are some of the related threads:
http://www.reality.com/dm/modules.php?name=Forums&file=viewtopic&t=2995&highlight=&sid=a4f2cc6ad98c3fa17f18fd3ca101bb05
http://www.reality.com/dm/modules.php?name=Forums&file=viewtopic&t=2970&highlight=&sid=a4f2cc6ad98c3fa17f18fd3ca101bb05
http://www.reality.com/dm/modules.php?name=Forums&file=viewtopic&t=2952&highlight=&sid=a4f2cc6ad98c3fa17f18fd3ca101bb05
http://www.reality.com/dm/modules.php?name=Forums&file=viewtopic&t=2421&highlight=&sid=a4f2cc6ad98c3fa17f18fd3ca101bb05
http://www.reality.com/dm/modules.php?name=Forums&file=viewtopic&t=2215&highlight=&sid=a4f2cc6ad98c3fa17f18fd3ca101bb05

Unchartered Poster Joined: Aug 02, 2011 Posts: 18

Woody wrote:

I delved into this issue a few years ago.
After data collection and reverse-engineering, there was one model for a rollup generator which matched the available data best.

Step 1: roll for size, normally distributed with a mean of 12 and a standard deviation of 4.
Step 2: randomly allocate 1 point at a time to the remaining stats until there are 70 points total.

I do not know what method the actual rollup generator uses for exception handling when a stat ends up with a value less than 3 or more than 21.
These happened about 1% of the time using this algorithm.

I could see it working this way, I could also see it working with three 7-sided dice rolled for all the other stats, then whatever is the remainder being lumped into Siz. On some rare occasions, it may result in a size > 21 or <3 ... and that setup could be discarded or somehow balanced by shaving off in one area and adding to another.

For instance, in my statistical data, it showed a slightly disproportionate strength stat and a heavily disproportionate size stat. Maybe if size < 3, they take some from strength to balance it - or maybe they just work at the top of the list (strength) and work down from there taking 1 from each in turn to add to Size to bring it up to acceptable margin of 3. And maybe if size > 21, they shave some off size and add it back into the other stats starting from the bottom of the list (Deftness) and working up?

There's probably a dozen ways they could do it, but no matter how you look at it, Size is the only stat not treated like the others in their algorhythm.

Posted: Mon Sep 26, 2011 6:10 am

Keeping tabs still (and adding my own rollups as I get them). I added the median and mode, no surprise on the median, but the mode seems agreeable to my intuition as to what it has always felt like.

9.34 9.78 11.94 9.76 9.58 9.85 9.72 average
10 10 12 10 10 10 10 median
9 10 10 10 9 9 10 mode

Posted: Tue Oct 18, 2011 12:29 pm

ImpInTraining wrote:

Woody wrote:

I delved into this issue a few years ago.
After data collection and reverse-engineering, there was one model for a rollup generator which matched the available data best.

Step 1: roll for size, normally distributed with a mean of 12 and a standard deviation of 4.
Step 2: randomly allocate 1 point at a time to the remaining stats until there are 70 points total.

I do not know what method the actual rollup generator uses for exception handling when a stat ends up with a value less than 3 or more than 21.
These happened about 1% of the time using this algorithm.

I could see it working this way, I could also see it working with three 7-sided dice rolled for all the other stats, then whatever is the remainder being lumped into Siz. On some rare occasions, it may result in a size > 21 or <3 ... and that setup could be discarded or somehow balanced by shaving off in one area and adding to another.

For instance, in my statistical data, it showed a slightly disproportionate strength stat and a heavily disproportionate size stat. Maybe if size < 3, they take some from strength to balance it - or maybe they just work at the top of the list (strength) and work down from there taking 1 from each in turn to add to Size to bring it up to acceptable margin of 3. And maybe if size > 21, they shave some off size and add it back into the other stats starting from the bottom of the list (Deftness) and working up?

There's probably a dozen ways they could do it, but no matter how you look at it, Size is the only stat not treated like the others in their algorhythm.

The data used for size were unpolluted/serial/without omission replacement rollups provided by Consortium and a couple others.
The 3d7 method does not give distribution consistent with data.

The test runs would go about 50k to 60k rollups at a time.
This was well beyond the point where frequencies stabilized.

It's entirely possible that the replacement rollups also come from a finite pre-generated pool.
The method used to generate that pool is one path to predicting the frequency of certain desired rollups.

Most of the long-term players and big tourney players long since acquired an accurate "gut" feeling for these frequencies.

Posted: Tue Oct 18, 2011 12:41 pm

[quote="Woody] It's entirely possible that the replacement rollups also come from a finite pre-generated pool.
[/quote]

I can believe that... how often do we see duplicate rollups? I've personally gotten my own clones more times than I can count, and at least once a month, I see someone post a rollup here that I've gotten before. What are the odds? No really, I'm asking what are the odds, I don't have the will power to crunch the numbers myself, but what is the mathematical probablility of getting the same rollup twice? I know there would be variables, let's just say if you get 50 rollups a year.

Posted: Tue Oct 18, 2011 1:58 pm

_Buri_ wrote:

[quote="Woody] It's entirely possible that the replacement rollups also come from a finite pre-generated pool.

I can believe that... how often do we see duplicate rollups? I've personally gotten my own clones more times than I can count, and at least once a month, I see someone post a rollup here that I've gotten before. What are the odds? No really, I'm asking what are the odds, I don't have the will power to crunch the numbers myself, but what is the mathematical probablility of getting the same rollup twice? I know there would be variables, let's just say if you get 50 rollups a year.[/quote]

If they are generated newly each time, it would depend on the particular rollup.
Not all rollups occur with the same frequency.

If it's a pre-generated pool...
Then, assuming no duplicates in the pool, it would depend upon the size of the pool.

Given some time and specifics, I can calculate the answer to the first question.
The second isn't knowable without knowing the size of the pool.

(Though if it is a finite pool, the frequency of recurrences could be used to calculate the size of the pool...)

Posted: Tue Oct 18, 2011 2:02 pm

Woody wrote:

_Buri_ wrote:

[quote="Woody] It's entirely possible that the replacement rollups also come from a finite pre-generated pool.

I can believe that... how often do we see duplicate rollups? I've personally gotten my own clones more times than I can count, and at least once a month, I see someone post a rollup here that I've gotten before. What are the odds? No really, I'm asking what are the odds, I don't have the will power to crunch the numbers myself, but what is the mathematical probablility of getting the same rollup twice? I know there would be variables, let's just say if you get 50 rollups a year.

If they are generated newly each time, it would depend on the particular rollup.
Not all rollups occur with the same frequency.

If it's a pre-generated pool...
Then, assuming no duplicates in the pool, it would depend upon the size of the pool.

Given some time and specifics, I can calculate the answer to the first question.
The second isn't knowable without knowing the size of the pool.

(Though if it is a finite pool, the frequency of recurrences could be used to calculate the size of the pool...)[/quote]

This is giving me a headache just thinking about it Shocked