Ranking Conditional

[dr. carbon]

To prevent people who just started from having an extremely volatile and inaccurate ranking, why not create a minimum game number to exceed to be ranked.

If we make it like 30 or something, this will balance out our ranking system, while still allowing the user to view his rating.

Any comments?

[gumbyy]

Originally Posted by dr. carbon

To prevent people who just started from having an extremely volatile and inaccurate ranking, why not create a minimum game number to exceed to be ranked.

If we make it like 30 or something, this will balance out our ranking system, while still allowing the user to view his rating.

Any comments?

I think this is a great idea. Anything less than x number of games (I'd say 20-40 minimum) doesn't give an accurate reflection of skill. I also think there should be some sort of minimum # of games / month or something like that. That way, someone who was really good 6 months ago but hasn't played in months wouldn't keep a top rating that doesn't necessarily reflect their current skill.

[elxir]

i think black ops does this and it seems like a good idea

[tantalum]

Gumby, you're bringing in a second issue - whether inactive vets should be allowed to retain their ranking. That's irrelevant to OP's post.


Regarding the volatility of new people, I've been thinking that for someone who's new on ladder, their ratings should go up or down by 100 for their first ten games, and thereafter, change by the normal amount. This way, they reach their equilibrium ranking a lot more quickly. This also breaks the zero-sum property of ladder, but I think that's a feature that's almost useless, since we have one-time visitors to ladder who do nothing but inflate the total pool of points in ladder by 25 points.

[Stormich]

All of this is easy to discuss, but without giving actual formulas nothing is really brought here. The current ladder balancing calculation has about 300 lines of code FYI

[lemon]

Being a new player to ladder I can definately say I got waaaaaaaay overrated at the beginning and still am despite loosing tons of games. How about giving different weight to rating changes for the first 20-40 games so that your elo drops/raises quicker?

[Stormich]

Most people agree that weighting initial games is the way to go, it's still really difficult to get it coded

[dr. carbon]

K first, storm, putting in a ranking conditional, i am sure, would be very little work for nobo. However, it would, in fact, add to his list of work.

First, idk how ladder code runs, but its very simple:

Create a boolean:
If gamecount >30 then Rankt (ranktruism) = true else Rankt = false
If Rankt = true then run Rank (procedure)


Simple solution to a complex problem.

Gumby's would only require what? 1 extra line of coding.

Now I am aware that I do not know the full brunt of the burden, but I am trying to say that it is very feasible.

[tantalum]

The intuitive way that the code should work is that after each game, it runs a script updating each player's score (i.e., the ranking update is game-event-triggered, rather than activating for individual players). I'm guessing it cycles through player names and runs an update function, passing it the parameters of +/- 24 or 25. (at least, this update function should be encapsulated rather than hard-coded in.) Then, in the update function, add a conditional that says,
if playergames < 10{
if scorechange > 0;
scorechange = 100;
else if scorechange <0;
scorechange = -100}

[dr. carbon]

Yea, seriously, if you want a +/- 100 range for up to 30 games, then follow tantalum's formula. If you want ranking to occur after 30 games, follow mine.