Suggesting new ELO system. Simple and concrete method.

[taywon2]

Hi everyone,

I was wondering if we could possibly have new ELO system. Current system uses ratio of total summation of each player's elo score. Well known problem of this method is that it doesn't reflect the hero dynamics of the team as well as k/d/a ratio.

In order to consider that, we should give separate scores for each players. Concept is very simple, we could give weight on each player who played for each game. We could easily find truth scores/weights for each player by following steps:

Key Steps:
1. let a1,a2~a5 are the heroes in the sentinel team and b1,b2~b5 are for scourge team.
2. find the exact game in the past where players played a1,a2~a5 and b1,b2~b5.
3. construct distribution of k/d/a for each heroes (due to CLT & LLG, this should be normal distribution)
3. normalize the distribution (z-score)
4. use mean and standard deviation to find the actual score.

Imagine if supporter did super well in the game but have less than 1.0 kda in the game. The new elo system will capture this by looking into past data and give more score to him. For example, if one played ES and did 2/4/30 in 35min and the other did 5/0/0 lina in 35min, ES probably gets more elo points.

I'm a phd student and have very much experience in data science, statistics, statistical/machine learning, econometrics, computer science etc. If we were to implement this system, I would be happy to volunteer to jolt down and start coding (i can do this in a day)

What do you guys think?

PS]I think this is the system that 11 (Chinese dota server) uses and I believe this system is very fair compare to the one that ent is using now.

[lodmaster]

Although I strongly support improvements in the ELO system, I'm against using k/d/a as a primary weighting factor because it will incentivize selfish behavior e.g. kill-stealing, fountain sitting (when losing).

Imo, the biggest weakness, by far, in the current ELO system is that players can make new accounts. This introduces constant inflation into the system and renders ELO to be meaningless for 90% of accounts. Just join a pub, and you'll see the vast majority are new accounts.

P.S. I'm not sure k/d/a would be normally distributed. Some heroes have different playstyles with distinct k/d/a outcomes e.g. playing an intel hero like lina as a support vs mid core may lead to a bimodal distribution.

[lodmaster]

P.P.S. Your profile says you live in Cambridge -- are you at MIT? I'm an alumni

[WineofViolence]

In the presence of the ability to make new accounts, the only thing to do is to adopt some measure of player skill that stabilizes relatively quickly.

ELO is not such a measure. Pretty sure simple win% is a better metric.

[lodmaster]

Win Pct seems like a good start as far as a metric that stabilizes quickly.

[EdgeOfChaos]

Winrate is far worse than ELO and will cause even further crazy stacking.\

There's no "correct" system because it is impossible to gauge the skills of a player from stats alone. ELO gets reasonably close, though there are flaws. Going to KD or Win% takes it even further away.

[Letbell]

Same here, I'm strongly against k/d/a scores - It's basically the same elo system that DotaVoid was used before Blizz screw up the bots.

To make it more simple, players gain/lose 25 elo for every matches that they played - Just like Dota 2 mmr ranking.

[kunkka]

Same here, I'm strongly against k/d/a scores - It's basically the same elo system that DotaVoid was used before Blizz screw up the bots.

To make it more simple, players gain/lose 25 elo for every matches that they played - Just like Dota 2 mmr ranking.

So stackers get 25 elo instead of 1 elo per game?

[EdgeOfChaos]

Yeah, that's not a good idea at all. You probably won't come up with a better system than ELO, sadly. Other systems, such as dota void's are pretty bad (funfact: I made dotavoid top ten in 30 games with some funny abusive builds, then stopped playing there)

[Sylvanas]

Void's score system was mostly based on individual game stats, regardless of game being won or loss. It compared how well you did to the average stats of the hero used and increased or decreased your score. I don't think it took account of the strength of teams, but it didn't punish you for failing to carry an uncarriable team or reward you for being carried while doing nothing useful. Of course you always tend to get better stats in better teams though, but if you were the kind of people that just follows strong players and let them do all the work, you weren't really going up in rank.

[EdgeOfChaos]

The method of champions was to go around killstealing with your preferred item or skill, then fountain sit when your team starts to lose. However I just abused a super broken hero (at the time, Earth Spirit) and would get like 20+ kills each game.

[Letbell]

Same here, I'm strongly against k/d/a scores - It's basically the same elo system that DotaVoid was used before Blizz screw up the bots.

To make it more simple, players gain/lose 25 elo for every matches that they played - Just like Dota 2 mmr ranking.

So stackers get 25 elo instead of 1 elo per game?

But then...both team would either gain/lose 25 elo =P
It's pretty simple... it's either you win or lose, that's all.

Stackers that stack pixel numbers doesn't mean anything to be honest.

PS: the worst part is that pub players doesn't communicate.
I know this isn't like wc3areana or something like Dota 2 where you can have voice-chat that communicate with other people. Having a voice-chat party can greatly improve gaming experiences, but it's just that people are here for casual dota gaming so yeah.

Most pubs people fear stackers, because they (pub players) are doing solo queue with pubs that doesn't communicate, resulting a bad game for them.
If you noticed, that people that stack in a game do communicate and understand each other (and heck they may even use discord or some form of communication application) too where as solo queue player does not.

DotaVOID elo system is based on individual's skill where as ENT elo system is based on team's performance.
If you are playing as a solo queue in a pub game then you'll expect the expected (either good or bad gaming experiences).
If you are playing as a party in a pub game then you'll expect to have a decent game.

This is like...you play what you get for.
Play solo with some random pub with no communication = bad/good game depend on how well you punish enemy's mistakes
Play with a party that has a form of communication = always a decent/good game

If you want a good example of a mix party + solo queue: https://entgaming.net/findstats.php?id=11589795

This is a game where my buddy and I playing 2v5.
Three players on our team refuses to do anything while just farming & griefing because they don't know how to deal with their own lane even though they have the best match up.
Notice sven attempt to feed meepo on purpose, sudicide on first blood and attempt to do it again and again. My buddy and I didn't even bother communicate with him, because he's doing it on purpose, and there's nothing we can but following our roles.

In the end, we won the game by punishing the enemy team's mistakes.

1) Meepo rushing dagger and doesn't gank, and we were like "Wtf is that meepo buying?"
2) The enemy team doesn't communicate well and doesn't understand to take advantage of the situation when sven is feeding on purpose.

PS2: My suggestion that ENT should force both casual and non-casual players to have some sort of voice-chat so that they understand each other both objectively and gaming experiences. Although ENT dota gaming is for casual players, and there's no competitive involved. It's either you win/lose or enjoy/not enjoy it.

[EdgeOfChaos]

What the fuck are you going on about dude... That's an incomprehensible mess.

[Letbell]

I'm drunk.

[Sylvanas]

I'm drunk.

Since December 1st?