DotA Match-Making

WineofViolence
Resource Storage
Posts: 12
Joined: Thu Jun 13, 2013 4:25 pm

Re: DotA Match-Making

Postby WineofViolence » Mon Aug 13, 2018 6:52 pm

The new system seems promising, but I'd really ask that you use KDR rather than ELO for the balancing, or at least factor KDR in to some degree. KDR "converges" much more rapidly than ELO to a value representative of a player's ability, which greatly limits smurfing ability. Obviously, it is not a perfect measure - a skilled player who tends to pick support heroes will tend to be undervalued by KDR and a mediocre player who picks carries will tend to be overvalued - but it is on the whole a far better metric.

EdgeOfChaos

Re: DotA Match-Making

Postby EdgeOfChaos » Tue Aug 14, 2018 5:25 pm

tbh matchmaking isn't a great idea anyways just because of the ability to make unlimited smurfs, it's never gonna be accurate :\

WineofViolence wrote:The new system seems promising, but I'd really ask that you use KDR rather than ELO for the balancing, or at least factor KDR in to some degree. KDR "converges" much more rapidly than ELO to a value representative of a player's ability, which greatly limits smurfing ability. Obviously, it is not a perfect measure - a skilled player who tends to pick support heroes will tend to be undervalued by KDR and a mediocre player who picks carries will tend to be overvalued - but it is on the whole a far better metric.


alright time to bring out the old dagon rikimaru build, I can get high kd therefore I am pro

WineofViolence
Resource Storage
Posts: 12
Joined: Thu Jun 13, 2013 4:25 pm

Re: DotA Match-Making

Postby WineofViolence » Tue Aug 14, 2018 8:56 pm

EdgeOfChaos wrote:alright time to bring out the old dagon rikimaru build, I can get high kd therefore I am pro


Obviously, there are ways to "game" the system no matter what metric you choose; I'm still pretty sure KDR-based matchmaking will do a better job on the whole of balancing games than Elo-based matchmaking, given how easily people can make smurfs.

Alternatively, you could use straight-up win percentage. Or, for an even more-radical take, you could just randomize the teams, which is potentially-frustrating but impossible to "game" in any capacity.

EdgeOfChaos

Re: DotA Match-Making

Postby EdgeOfChaos » Tue Aug 14, 2018 11:50 pm

i mean
a smurf has unknown KD too.. just like unknown ELO.

Sylvanas
Treant Protector
Posts: 974
Joined: Wed Jun 25, 2014 4:56 am
Been thanked: 159 times

Re: DotA Match-Making

Postby Sylvanas » Wed Aug 15, 2018 12:27 am

EdgeOfChaos wrote:i mean
a smurf has unknown KD too.. just like unknown ELO.

He's saying it takes only a few games for the smurf's "real" approximate kdr to reveal itself, while it takes many more games for them to be given a fair elo. If you're 1500 elo and 2 kdr or 500 elo and .5 kdr and you play on a smurf/new account as you normally would, your kdr will usually be representative of who you are as a player after 5-10 games, while elo still has a long way to catch up. Yes, you can inflate your own kdr at the expense of your own team by doing selfish dagon shitbuilds, but you could argue that people can inflate their elo as well. I mean, just look at that link in your signature right now...

It's impossible to create a perfect system as long as we're dealing with b.net and its disposable, created in 10 seconds accounts, but I think it's a step in the right direction.

EdgeOfChaos

Re: DotA Match-Making

Postby EdgeOfChaos » Wed Aug 15, 2018 12:47 am

that makes no sense, you can be a trash player and get a 10 KD in a game

Sylvanas
Treant Protector
Posts: 974
Joined: Wed Jun 25, 2014 4:56 am
Been thanked: 159 times

Re: DotA Match-Making

Postby Sylvanas » Wed Aug 15, 2018 1:06 am

EdgeOfChaos wrote:that makes no sense, you can be a trash player and get a 10 KD in a game

One game, not 3, even less 5 or 10. And more often then not, a trash player will get a trash result in a game either way. No system (or lack of one) can use a single game to rank players. Why are you rebutting a point no one made?

User avatar
kunkka
Treant
Posts: 442
Joined: Tue Jan 12, 2016 7:53 pm
Has thanked: 84 times
Been thanked: 50 times

Re: DotA Match-Making

Postby kunkka » Wed Aug 15, 2018 1:30 am

Well no point arguing about kdr on smurf account here. The first game of my last new account, I got dumped to a game with average 750 elo on both sides and that was a shit show. Won't go apem with new accounts any more rofl

EdgeOfChaos

Re: DotA Match-Making

Postby EdgeOfChaos » Wed Aug 15, 2018 1:40 am

Sylvanas wrote:
EdgeOfChaos wrote:that makes no sense, you can be a trash player and get a 10 KD in a game

One game, not 3, even less 5 or 10. And more often then not, a trash player will get a trash result in a game either way. No system (or lack of one) can use a single game to rank players. Why are you rebutting a point no one made?


hmm a point no one made AreYouSure
let's go back with the ol quote tool shall we

I'm still pretty sure KDR-based matchmaking will do a better job on the whole of balancing games than Elo-based matchmaking, given how easily people can make smurfs.

smurf = brand new account
kdr will be wildly inaccurate for smurfs, far more inaccurate than ELO.
because smurfs can reach very inaccurate KD in a few games, and yes it can still be extremely inaccurate in like 5 games, or even 10
yes it becomes more regular after you play 10-20 games but we're talking about smurfs here, theyll make a new acc at that point
therefore it does exactly the OPPOSITE of what he said. It is a worse system for smurfs.

Sylvanas
Treant Protector
Posts: 974
Joined: Wed Jun 25, 2014 4:56 am
Been thanked: 159 times

Re: DotA Match-Making

Postby Sylvanas » Wed Aug 15, 2018 2:03 am

A small sample size is better than nothing. elo is nothing until you have played enough games for it to stabilize, and that takes dozens of games. Smurfs obviously don't have much information to manage with, but you have to take what you can get. Unless you go out of your way to skew your own stats on purpose, your real kdr should show after a few games, and should only get more accurate after more games (even if actively trying to distance yourself from it). "Brand new" 0 game smurfs are irrelevant to how people should be ranked, because you can't use any statistic they have, since they have none.

WineofViolence
Resource Storage
Posts: 12
Joined: Thu Jun 13, 2013 4:25 pm

Re: DotA Match-Making

Postby WineofViolence » Thu Aug 16, 2018 5:18 am

Sylvanas wrote:A small sample size is better than nothing. elo is nothing until you have played enough games for it to stabilize, and that takes dozens of games. Smurfs obviously don't have much information to manage with, but you have to take what you can get. Unless you go out of your way to skew your own stats on purpose, your real kdr should show after a few games, and should only get more accurate after more games (even if actively trying to distance yourself from it). "Brand new" 0 game smurfs are irrelevant to how people should be ranked, because you can't use any statistic they have, since they have none.


This.

KDR and win percentage both stabilize much faster than Elo. They might not be as unbiased an estimator of player quality as Elo is, but Elo takes far too many matches to stabilize be useful as a statistic in the presence of smurfs. On the whole, they will be more accurate by virtue of working better on smaller sample sizes.

Of course, the "randomize the teams" option is also perhaps a good one ;)

EdgeOfChaos

Re: DotA Match-Making

Postby EdgeOfChaos » Thu Aug 16, 2018 7:10 pm

small sample size is not better when it can bring you to a very wrong conclusion
https://entgaming.net/openstats/dota/?u=dotarobot
here, I just played 3 games, didn't even use meepo, got a 14.7 to 1 KD. wasn't even stealing kills or anything
my real KD on account is 2-3 so rating me on kd, the game thinks i am about seven times better than I really am. KD system is VERY sensitive to smurfs.
not to mention that KD isn't a good indicator of skill even when it stabilizes, like you can be a great support player and have a KD under 1
Attachments
uh-oh.PNG
ez best player on ENT
uh-oh.PNG (6.41 KiB) Viewed 3101 times

WineofViolence
Resource Storage
Posts: 12
Joined: Thu Jun 13, 2013 4:25 pm

Re: DotA Match-Making

Postby WineofViolence » Thu Aug 16, 2018 9:53 pm

Your example is quite bad, actually; your "ordinary" k/d ratio is already towards the upper-end of the bell curve; all one needs to do is use a resistant statistic (quantile rather than value) and you haven't disguised anything at all.

I think it's pretty ridiculous to suggest that Elo is more accurate than KDR or win% given the average number of games played by an account on these servers.

Sylvanas
Treant Protector
Posts: 974
Joined: Wed Jun 25, 2014 4:56 am
Been thanked: 159 times

Re: DotA Match-Making

Postby Sylvanas » Thu Aug 16, 2018 10:19 pm

EdgeOfChaos wrote:small sample size is not better when it can bring you to a very wrong conclusion
https://entgaming.net/openstats/dota/?u=dotarobot
here, I just played 3 games, didn't even use meepo, got a 14.7 to 1 KD. wasn't even stealing kills or anything
my real KD on account is 2-3 so rating me on kd, the game thinks i am about seven times better than I really am. KD system is VERY sensitive to smurfs.
not to mention that KD isn't a good indicator of skill even when it stabilizes, like you can be a great support player and have a KD under 1

You're just nitpicking. Where are all your other smurfs from when that didn't happen? Where are all of the innumerable <1 kdr support players in the top players list? You've previously mentioned terrible players that risk getting 10 kdr in their first game on a new account, care to show me some of those? You're just pointing out the statistical anomalies that are inevitable considering those small samples you have to work with. Those few games aren't enough to prove or establish anything, but they're enough to make an educated guess. You're basically telling me that a rock is better than a pencil for writing because the tip of a pencil can sometimes break.

15 kdr over 3 games means good. Not likely to be 15 kdr good, since no one can maintain anything near that, but you can safely assume it's not b.e.a.s.t's smurf, so you could have that player play in the higher ranked games. Since the playerbase isn't huge and it's not like 10 kdr and over exclusive games would be created either way, judging that account by its kdr would be perfectly effective for match-making purposes. Were you worried you'd be matched with nine different 10+ kdr players or something?

On the other hand, judging it (and every other smurf, good or bad) by their elo would basically rank them all as "average player of approximately 1000 elo" and achieve absolutely nothing. ELO just isn't made to rank people accurately until enough data has been compiled. If what you have is 3 games, it's best to focus on what those 3 games were like.

User avatar
kunkka
Treant
Posts: 442
Joined: Tue Jan 12, 2016 7:53 pm
Has thanked: 84 times
Been thanked: 50 times

Re: DotA Match-Making

Postby kunkka » Fri Aug 17, 2018 7:19 pm

uakf.b wrote:We are beta testing match-making system on the apem bot.

The beta test will not be happening all the time, just for a few hours. We will change things based on feedback, please reply below!

Are we move back to match-making any time soon? Thinking to have a main account again :)


Return to “Archive”

Who is online

Users browsing this forum: No registered users and 7 guests