Ranking system

[Lopdo]

We can easily counter excessive growth of rank for high ranked players by introducing same mechanics that HoN uses (probably lot of other games as well, but this is only game I played), e.g. more points you have, less points you gain/lose, if you have 1000 points, you can win/lose 20-40 points, if you have 2000 poitns you can win/lose only 2-4 points.

I am not sure if Jorion's formula implemented this already but I don't think so

[Joriom]

Since we all know how much we're all obsessive gamers and how much time we're going to spend on this, 8 players with 100 000 points; the winner will receive 778 points, which is an increase of roughly 10 000% from the initial winning value.

You're right. I've thought about it before but I've taken wrong assumptions which did lead me to wrong conclusion. My main idea was to have fixed grain value for winning against opponents with same ranking. It should stay same for 8 players with 1000 and 200,000 rating and not proportional to their rank value. I've got both variants coded but I've messed something with the GOOD one and I've decided just to comment it. If you look at my webpage source code you'll see big commented block with "Wariant: oponents based cost, softcore* gain" header

We can easily counter excessive growth of rank for high ranked players by introducing same mechanics that HoN uses (probably lot of other games as well, but this is only game I played), e.g. more points you have, less points you gain/lose, if you have 1000 points, you can win/lose 20-40 points, if you have 2000 poitns you can win/lose only 2-4 points.

I am not sure if Jorion's formula implemented this already but I don't think so

You've just send it while I was writing my post now. I've just realized that my formula does exactly the opposite to what you say. Although I can't agree with DECREASING the gain ratio. When I fix some calculations it should create artificial cap based on number of accounts regularly playing. At least I hope it will.

you take the average of their points so if u've played one game and you took 1/8 you have the highest score possible and can only go down.
because that would really mess up the leaderboards just make this stat not display until they've had 50 games.

Sorry, I did misunderstood "mean". You could even make it in range 0-100 pts so it would look like percent later which would be nice. But It still does not include one important for me variable. Opponents strength. Winning ten times against newbs would give same results as winning 10 games against techgumps

[BelgarionRiva]

I have a suggestion, unlock special features as points grow.

Example:
EDIT: By rank, I mean points.
Once rank reaches 1050, unlock background (unlocked forever)

Once rank reaches 1200, unlock race with stats we already know, but cool addition (unlocked forever)

Once rank reaches 1500, unlock xp upgrade +5% (unlocked as long as rank remains over 1500)
At this point, it will be an incentive to stop playing ranking matches, but by putting more rewards, the incentive to keep playing ranked matches, and thus to get premium for life, greatly increases.

Once rank reaches 2000, unlock shards upgrade +5% (By then, players that reach this point probably already have all the upgrades, so offering more shards does not actually matter; but to new players, it is a great incentive to get premium for life.)

Keep adding badass upgrades, etc, etc, etc.

I'm not sure what you mean by this? give a reward to people who are doing good so that they can do better easier?
i think bragging rights is better.

Fine, bragging rights, races with no bonuses, stuff that looks cool but does not much else. Better? I thought that giving people bonuses for being good was an incentive to try to be better...

[egavaS]

you take the average of their points so if u've played one game and you took 1/8 you have the highest score possible and can only go down.
because that would really mess up the leaderboards just make this stat not display until they've had 50 games.

Sorry, I did misunderstood "mean". You could even make it in range 0-100 pts so it would look like percent later which would be nice. But It still does not include one important for me variable. Opponents strength. Winning ten times against newbs would give same results as winning 10 games against techgumps

that is true, however that should even itself out with 50+ games played.

[egavaS]

Fine, bragging rights, races with no bonuses, stuff that looks cool but does not much else. Better? I thought that giving people bonuses for being good was an incentive to try to be better...

people already want to be the best in this game, no one plays hoping for second or third or penultimate.

[KivoR]

We can easily counter excessive growth of rank for high ranked players by introducing same mechanics that HoN uses (probably lot of other games as well, but this is only game I played), e.g. more points you have, less points you gain/lose, if you have 1000 points, you can win/lose 20-40 points, if you have 2000 poitns you can win/lose only 2-4 points.

I am not sure if Jorion's formula implemented this already but I don't think so

?????
What is the purpose of this ranking system ? I believe it is to evaluate the strengh of the players. A player may wish to improve and get better results. But the system should only reflect the results. The system should compute that strength rating as fast as possible match after match. There is no reason to slow down "an excessive growth" !!! If people (good and bad ) have growing ratings, if the system only inflates,... then it fails at evaluating the players. If the system doesnt give a stable rating to a player then it fails.
You may slow down the variations of the rating of a player after a while. But only if you are considering you reached the value of his strength and that you want to reduce random variations of his ranking.

If this ranking system is only about gathering points and points. Then maybe it is just a race for points where strong ones go faster, early players started first. And maybe you want to regulate it to make it more attractive. But that is a very different philosophy.

According the presented formula, I believe this evaluates the strengh of the players (their smart and their boosts).

[KivoR]

people already want to be the best in this game, no one plays hoping for second or third or penultimate.

I see a problem. 50 % of the players are under the average ! ! !

They begin with 1000 points like anybody then they only go down. They can t rely on noobs because they are not rated. And when noobs are rated, they are not noob anymore after a while.

One way to fix that is to have hidden points and shown points.
You start with 0 shown points and 1000 hidden points.
When you win, you get shown points.
When you lose, you lose hidden points.
You compute the matches with the points wich are the sum of both hidden and shown points.
When you have no hidden point left you go on with only shown points (winning and losing them). At that point your rating have been only climbing. You may consider you reached a fair evaluation of your rank (but you can still improve).

That makes the formula more complex, needs a variable that will be useless after a while. I have no better idea to fix it.

[Joriom]

Ta-da-dam! Formula updated!

New version of formula is up and running! I won't even try to explain how it works exactly. Until it proves to be worth it at least. It's just too much complicated at this point. Also I've added some stuff to help testing. Configurable randomizator and possibility to pass results as base for next iteration. So... go and play with some values! And don't forget to post here is you see any problems!

http://joriom.site90.net/vw/ranktest.php

For curious I'll post part of my PHP code used for calculations:

Code: Select all

$MIN_RANK = 500;
$MAX_CHANGE = 0.1;
/* ONCE PER GAME */
$gsum = $pool = 0;
for ($i=1; $i <= $players; $i++)
         $gsum += $points[$i-1];
for ($i=1; $i <= $players; $i++)
         $pool += $gsum / $players / $points[$i-1];
$totalparts = (($players-1)*$players/2);

/* FOR EVERY PLAYER AT WIN/DEFEAT */
/* DRAW rules should be easy enough to implement
    They would need change only in $parts value */
$parts = ($players-$place);
$cost = ($players-1) * ($players) * $pool * $players * $points[$p-1] / ($gsum * $totalparts * 2);
// if maximal $gain would be lower than $cost we do recalculation
if (($players-1) * $players * $pool / $totalparts < $cost)
   $gain = $cost * $parts / ($players - 1);
else
   $gain = $parts * $players * $pool / $totalparts
$change = round($gain - $cost);
// you can't go lowe than $MIN_RANK but also can't change rank more than % defined by $MAX_CHANGE
// $MAX_CHANGE is optional, matches with over 1000% rank difference were problem
// $MIN_RANK solves it as well. Its just in case $MIN_RANK would be low enough to cause problems
$newpts = max($MIN_RANK,$points[$p-1] + $change, round($points[$p-1] * (1-$MAX_CHANGE)));
$newpts = min($newpts, round($points[$p-1] * (1+$MAX_CHANGE)));
// Optional, for information purposes.
$change = $newpts - $points[$p-1];


Yes, to calculate $cost I'm using almost $players raised to the third power. Hell if I know how did I manage to come to that formula. It was based on some simple facts that I was constantly merging, shuffling and rolling over. I did not write down my reasoning. Some of few last steps were made by "idk what is it, but i hope it won't get mad if i change it" rather than "i should improve that and that". Warning! This code is not optimized in any way. I just didn't waste time on that as it may still be subject of change.

[Joriom]

As KivoR suggested I've made simulation of 1000 matches with fixed winning probabilities for every player to check how long would it take for ranking to get well set up and if it would stabilize at all.

Here are results:


I've also upgrades "visual effects" on the page with rating calculator, but you might need to use F5 / refresh page for the new CSS to load.
http://joriom.site90.net/vw/ranktest.php

[BelgarionRiva]

Ranking should be included in an eventual VW2.0