TOPIC: Proposed GT Class Lottery Approach

Tobin Blake wrote:

bpsymington wrote: ... and if there is, there can be two of us!

We paladins were all too aware of this. Thankfully, that's being rectified with this year's token set.

Yeah, baby!

Quack.

Raven wrote:

kurtreznor wrote: so, I have an idea/method...but it requires a neutral entity to make final decisions:
sign-ups should involve a list of your preferences, in order of importance. so, the following items will be listed with the selection that is most important to you first.
1. class preference (list a few options in preferred order)
2. combat vs puzzle
3. dungeon A vs B
4. difficulty
5. special request (ie. keep me with my spouse)

(snip)

then, at some agreed time (I am thinking July 1, to allow a month after event tickets are sold but still a month before gencon), the neutral entity assigns people their run and class to 'maximize happiness'.

I like the idea. I'm a little uncomfortable with the "neutral entity" though. Partly because there is no truly "neutral" entity. Even computer programs are only as neutral as their programming, which can be designed to favour certain factors, and people who know how the code works can game the system.

I'm not saying letting people sort it out themselves online would necessarily be better - if people don't want to make waves by requesting a class switch, they probably also won't want to make waves by saying "No, I'd really prefer to be in group A over B" if A is turning out to be a group of friends sorting themselves all into the same group.

I don't think we're going to find a perfect solution. Perhaps we can find one which makes a greater number of people happy, or one which is a little more fair to players signing up later on, but there will always be people who are frustrated by whichever system is used... and sometimes something as impartial as "first come, first served" or as well understood as "that's the way we've always done it" is not so unreasonable as it sounds.*

Perhaps we can try - for this year, at least - having people list a variety of preferred classes, and having a notice on the sign-up page that nothing is set in stone. See how it works, and make adjustment as needed for future years.



* This opinion brought to you by someone who has benefitted in the past from "the way it's always been done" and is aware of their bias in that regard.

I started thinking about this approach, but the total number of possibilities is so enormous as to defy my processing capacity.

Given 4 distinct runs of 10 players, each of which could have one of 12 classes there are (I think):

40! * (12 choose 10)^4 / ( 10!^4 * 4! ) = 3720124179722959651860497280 ways to assign players to classes and runs.

This means it's not going to be practical to find an optimal allocation.

It's probably possible to find a stable allocation (e.g. no two players would agree to switch their positions).

To be clear, something could be done here - this type of problem is solved every day by universities that build student schedules - but I don't think we could hope for optimality.

Raven wrote:

kurtreznor wrote: so, I have an idea/method...but it requires a neutral entity to make final decisions:
sign-ups should involve a list of your preferences, in order of importance. so, the following items will be listed with the selection that is most important to you first.
1. class preference (list a few options in preferred order)
2. combat vs puzzle
3. dungeon A vs B
4. difficulty
5. special request (ie. keep me with my spouse)

(snip)

then, at some agreed time (I am thinking July 1, to allow a month after event tickets are sold but still a month before gencon), the neutral entity assigns people their run and class to 'maximize happiness'.

I like the idea. I'm a little uncomfortable with the "neutral entity" though. Partly because there is no truly "neutral" entity. Even computer programs are only as neutral as their programming, which can be designed to favour certain factors, and people who know how the code works can game the system.

Well, the computer program will just need to be an advanced AI that has learned how to tell when someone is trying to game the system and respond by enslaving all of humanity...I mean, 'by not rewarding such behavior.'

When you have an NP Complete problem like this, one can use techniques such as Greedy Method, Simulated Annealing, Ants Algorithm, Reinforcement Learning, and other techniques to get solutions that are "good enough"...

I use to do these at my place of work, and even taught these techniques in college... and they are fun.

There are optimizing routes that can be used to simplify it from that large of a search space to make it smaller...

The question is finding an acceptable answer that is "good enough"...

Maybe using a genetic programming technique would be cool, if we can define a way to encode the chromosomes...

Just thinking of the fitness function is fun! Using genetic programming, we could design an algorithm and I can already see a way to encode the chromosomes... for four teams of 10 players with up to 12 classes, it would require six bits per each of ten classes times four teams for a total size of 6 * 12 * 4 bits in size, and have some rules on having each set of twelve genomes require two set at zero meaning that class is not used... then override the rule of where we split the chromosomes to require split points between and not in the midddle of the chromosomes. and for mutation, just implement that by "swapping" two genomes, with a rule that if one picked to swap is non-zero, swap with another that is non-zero or swap within a 12 set group ignoring the value...

The fitness function would require everyone to specify more information, such as:
Puzzle run vs combat
Three classes IN Order of preference
One or two classes that they DONT want to be (could still get assigned if increases fitness/happiness)
one or two players that want to be on same team
One or two players the don't want to be with

We could have some code to do smart swapping to improve chromosome qualities as well as just the random splitting and recombining the chromosomes...

Mutation can be implemented to support never allowing for invalid chromosome states, such as keeping two character assignments always zero ten player max per team with twelve character choices)

Maybe I have been drinking too much tonight, but the more I think of this, the funner it sounds implementing said optimization program

Better sign off before I sound like more of a geek/nerd (if that is posible?)

Einstein said, "Make everything as simple as possible, but not simpler."

IMO, what Raven suggested pretty much fits that description.

Also remember that "better is the enemy of good enough"... but that doesn't suggest settling for mediocrity...

Mongo wrote: Just thinking of the fitness function is fun! Using genetic programming, we could design an algorithm and I can already see a way to encode the chromosomes... for four teams of 10 players with up to 12 classes, it would require six bits per each of ten classes times four teams for a total size of 6 * 12 * 4 bits in size, and have some rules on having each set of twelve genomes require two set at zero meaning that class is not used... then override the rule of where we split the chromosomes to require split points between and not in the midddle of the chromosomes. and for mutation, just implement that by "swapping" two genomes, with a rule that if one picked to swap is non-zero, swap with another that is non-zero or swap within a 12 set group ignoring the value...

The fitness function would require everyone to specify more information, such as:
Puzzle run vs combat
Three classes IN Order of preference
One or two classes that they DONT want to be (could still get assigned if increases fitness/happiness)
one or two players that want to be on same team
One or two players the don't want to be with

We could have some code to do smart swapping to improve chromosome qualities as well as just the random splitting and recombining the chromosomes...

Mutation can be implemented to support never allowing for invalid chromosome states, such as keeping two character assignments always zero ten player max per team with twelve character choices)

Maybe I have been drinking too much tonight, but the more I think of this, the funner it sounds implementing said optimization program

Better sign off before I sound like more of a geek/nerd (if that is posible?)

Sorry Batman, your out. Time to get me a new t-shirt:

"Always be yourself, unless you can be Mongo. Then always be Mongo"....

Mongo wrote: Also remember that "better is the enemy of good enough"... but that doesn't suggest settling for mediocrity...

Sure it does, because of the Pareto Principle. (80/20 rule.) If you can achieve 80 percent of the benefit from the first 20 percent of effort, that means you can solve five problems at 80% efficacy with the same resources, time and patience it would take to solve one at 100% while leaving the other four untouched.

But, if you have infinite resources and just one problem, sure, solve it 100%, or at least to the point where the incremental cost equals the incremental value.

Plus, the old military adage : if the minimum wasn't good enough, it wouldn't be the minimum.

And here I thought that Zeno's Paradox showed you could never reach 100%... tha you must always complete half the remaining portion and that there are an infinite number of halves... so you can get infinitely closer to 100% but never arrive...

I have never heard that military adage before, but I really like it!

Mongo wrote: And here I thought that Zeno's Paradox showed you could never reach 100%...

I did say "infinite resources." And time is a resource, so...

I have never heard that military adage before, but I really like it!

There is great comfort in setting reasonable expectations.