forum.webdiplomacy.net

Okay I've had a look at this and made a spreadsheet to test it out


It's designed to stop inflation, and it'd definitely do that; as you get more points games will, on average, always return a less of what goes in. It's not really zero-sum, but it would make the average value converge to a fixed place.
The problem is I don't like this and see inflation as an important part of the points system. If someone makes it to the top and can only go down why continue to play (assuming you're playing for the ranking)?

If a bunch of big points players decide to play a game they're sacrificing a significant chunk of their points just to do so; most will lose out, which is a pretty clear disincentive to play


I like how you can specify how the points are distributed at the end, but it doesn't seem very flexible in practice: I can't get it to approximate PPSC, at best it's WTA with a survival bonus

Also the tweaking of parameters varies how much the game will return on average. The minimum R value always seems too close to the starting player. The game value is automatically increased, but the amount of the game value returned at the end doesn't add up

It's also pretty complex, and there are a lot of magic numbers. I think this system has lots of big drawbacks and its benefits are debatable

Kestas

Your post mentioned several points and I will address them in order:

Inflation
You should consider reassessing your position on this.
The only benefit of inflation is to promote more activity so players don't fall behind.
The biggest problem with this is apathy. At some point even the good players will not care about their rank any more because it's simply too much effort to keep up with the inflation, which is at least linear. The last thing you want is apathy towards ranks as this is the mechanism that defines your incentives.
I do not get rid of inflation completely, but merely where it hurts us. See below.

Zero-sum
I actually mentioned this already. It is indeed not zero-sum. I said at the beginning that it may be a mistake to restrict ourselves to zero-sum functions as there are many interesting incentives one can define when we choose other functions. The fixed point was designed to do precisely what it does in your simulation.
If you have already made a simulation, I would like you to try the following:
Assume that the total population is growing. This is after all the desired scenario.
New population of course starts at rank 0.
And assume that population skill distributes normally in some sense. This is important, as with a larger population you will see stronger players.
I think you will observe some interesting phenomenon:
1) The average rank of the population should converge to just below the fixed point, depending on your rate of growth.
2) The top rank will indeed grow, despite the negative sum on high-ranked games. This negative sum is in fact hardly noticeable, it only curbs the inflation of the entire population. What really stops the point race is the penalty for playing against lower ranks, and not so much the negative sum.
I see this as getting the best of both worlds. Good players can take a rest knowing that they won't fall behind too far, while active players know that with some work they can get to the top.

PPSC
"P=0,W=.55,S=0 is a good approximation of PPSC"
That is the winner takes 55% and the rest divided PPSC.
This is even better than PPSC. It has all the same benefits as PPSC, but removes the horrible flaw that is the incentive to maximise your win.
All wins are equal.

I think you misunderstood the parametrisation.
There is Participation bonus, Winner portion and Survival* bonus, and the rest goes Per SC to survivors* (*minus the winner).

Game value
There is a good reason that the minimum value is not far from the initial value.
Only 10% of the points were invested when we set G=.1
Players can optionally change G.
I don't see a problem here.
The default merely tries to keep the system stable so that players' ranks don't jump erratically.
The total game value doesn't add up by design.
That was the whole thing about not being zero-sum.
It has a strong positive sum for low ranks and weak negative sum for high ranks.
This achieves:
1) an incentive to participate for new players
2) fast convergence to appropriate rank
3) zero total inflation
4) negligible harm to higher ranks

Complex
Indeed.
It can be simplified tremendously either via interface or redesign.
By interface you can simply set some basic settings for the parameters and players choose (perhaps with an options to tweak).
The only parameter that is chosen freely by the players is G = % points invested.
Interfaces can range over (Default, WTA, "PPSC") X (Unranked,Friendly,Serious) giving 9 basic settings for the parameters that the players don't even have to know about.
Alternatively, we can go back to the basic principle, and redesign elements that you don't like.
The basic idea was to have a function that was globally "zero-sum", but locally that depended on the game rank, giving a boost to low ranks and a slight hindrance to high ranks. Also, I had other incentives in mind when I design the point distribution function. There are limitless variations.
Not sure what you mean by magic numbers.

Ultimately I think my system is at least as good as the current system.
There are some important differences however, and the real question is if these differences justify the effort of coding.
This is not for me to decide.
You understand the efforts of coding. It might be useful to gauge the importance of these differences to active players.

Here are the important differences to consider:
Pros:
1) No inflation at the average rank, creating a reliable gauge of player strength.
2) Some inflation at the top ranks in a growing population, affording all the benefits of the current system's inflation.
3) Stopping the phenomenon of rank by quantity. A player cannot rise to high ranks by playing only weaker players, because the function is weighted, not because of the skewing.
4) 2 & 3 together should reduce the apathy of veterans to the ranking system, while still maintaining a spirit of competition.
5) Extra incentives for new players, participation, and survival.
6) Better differentiation of win/loss/survival.
7) Fast convergence to correct rank.
Cons:
1) Marginally more complex for players
2) Significantly more complex for admin

If you don't see how these pros are achieved I can elaborate. I just don't want to make the post too long. It's already heavy.

I haven't really looked in on this forum for a long while- I have pretty much decided that I can accept the current situation where I self-calculate my own system everyone once in a while.

This thread has one excellent part, more valuable than any system. It is this:


Functional (Promote desirable activity)
* Reward taking over CD [L]
* Reward mingling [G]
* Reward regular participation [G]
* Make restarting with a new account undesirable [G]
* Reward joining the right number of games [G]
* Punish CD [L]
* Punish inactivity [G]
* Encourage activity [G]
* Inviting to newcomers [G]
* Enjoyable to compete in
+ Encourage play at equal level [L,G]
+ Encourage quality over quantity [G]
+ Flexible to player time-frames [G]

Accurate (Act as a fair and reliable yardstick, and conserve the spirit of Diplomacy within the game)
* Reward surviving [L]
* Reward winning [L]
* All games shouldn't be equal [L]
+ Stability - player ranks should converge to some number rather than jump around irratically [L]
+ Non-stagnation - players should be able to change their ranks if they indeed improve (or worsen) [L]
+ Fast convergence to correct rank [G]
+ Continuity - player rank differences should mean the same at all levels [L] I would say, rankings should have some tangible meaning. As Kestas has said there is no way of stating how good a player is by just a number. You need that number to mean something. To say Rait is a player of standard 50 means nothing, but to say he wins 50% of his games means alot.
+ Time invarient - player ranks should be independant of time or frequency of play [G]
+ Winner gain is constant, to avoid point maximization within a game [L]
+ Distinguish clearly between victory, survival, defeat and abandonment [L]
+ Result in the same rating regardless of choice of opponent-added.

I shall make another post to deal with the proposed system

On a first reading, one thing struck me, which was that there is no PPSC/WTA distinction here, but rather a sliding scale. I don't think that that would be healthy for the community. Would it be possible to convert this to run on the current game types.

As for the results of calculation, I cannot see any catastrophic problem just looking at the formulae, so would want to be able to look at a model and values. Kestas, would it be possible for you to send me the model you have?

Finally would be to run the same database dump through this and Ghost-rating, and compare them.

TheGhostmaker, thanks for taking the time to read my crazy proposal.

With regards to the sliding-scale criticism. This is solved easily with standard configurations.
The fact that WTA and PPSC exist does not mean that there do not exist a continuum of possibilities in between.
I do not suggest that we bombard the users with this continuum, simply that we can define more subtle options than simple WTA and PPSC.
(Optionally we can give the users the option of tweaking, meaning they can choose freely from the continuum, if they feel confident they know what they're doing.)

I also noted that true PPSC is awful, because of the winner's incentive to drag out the game. PPSC-with-fixed-win is much better.
WTA and this modified PPSC are two of the special configurations that I propose, there are others that are useful too.
Useful in the sense that they promote certain kind of behaviour. (I always go to our root goals, behaviour modification.)

I am in fact in the process of rewriting my proposal, in a more organised fashion.
I don't know if that is helpful or not.
How much of my proposal did you understand?

I think I got a fairly good idea of it, but in terms of accuracy it was rather hard to analyse because two things didn't come through:

1. What does a rank of a given value mean? With the Ghost-rating, there is an Expected result function, so any given rank has definite, tangible meaning.

2. Does this function mean that, no matter who you play, there will be the effect that ranks turn out the same?

The reason for highlighting these is that they are the two most major problems with points.

1. Higher = better.
3000 = average player.
I think that players will get a feel for the numbers with time.
The numbers themselves mean very little.
0 = newbie
3000 = fixed point

Or perhaps you're looking for this answer:
The bet each player places is proportional to his rank, rather than an equal bet independent of rank.
This means that high-ranking players can't earn points off tons of low-rank games.
This is my #1 criticism of the current system.

2. No. This quality is called zero-sum, and my system does not have this. I did this intentionally and this is what I referred to as skewing.

I gave large positive sum for low-ranking games, for many reasons, like incentive to participate for new players and fast progress for good players so they don't have to play tons of weak players before they get to their level.

I gave slight negative sum for high-ranking players so that the entire system does not inflate. 3000 is the rank where games are zero-sum, hence I call it the fixed-point.

You can also check out my new thread on this system which tries to explain it better: A top down approach.

What I meant by 2 was not is it zero sum, but is it the case that no matter what standard of player a person plays, they will end up with the same rating?

I answered this under point 1.

"The bet each player places is proportional to his rank, rather than an equal bet independent of rank.
This means that high-ranking players can't earn points off tons of low-rank games.
This is my #1 criticism of the current system."

If you want a high you you really need to play high ranked players.
This makes the system a reliable form of comparison.