72 replies to this topic

[Revis Volek]

RussianWolf, on 30 December 2015 - 12:28 PM, said:

ELO was designed to target a 50% win/loss averagea dn was pretty close usually.

PSR isn't designed with that in mind as it has been explained. I just grabs 24 players of the 3 active tiers and goes. So you could see wider ratios in the W/L column.

ELO is designed to put two players of equal skill against each other...as far as i know w/l may be factored but ELO itself is not trying to give you a 50/50 its giving you a 50/50 chance at winning or close to as possible for the best skill match against players.

PSR works much the same, just because we got PSR doesn't mean the MM got thrown out the window. We still have a MM, what you are suggesting is NO MM which is false. Thats how CW works yes but not the normal queues.

[Tarogato]

Lukoi Banacek, on 30 December 2015 - 12:21 PM, said:

Taragoto, are you only sampling those dots or do they represent clusters of players? My KDR after 11000 matches is a full point higher than my win-loss and I pug about 80/20 (with statistically similar results). 2.59 and 57.1% win/loss. I am not sure in a large sampling size that KDR and win/loss would actually be similar.

Sampled individual players.

http://mwomercs.com/...-stats-results/

[sycocys]

Bilbo, on 30 December 2015 - 12:33 PM, said:

Again, if this were the case I'd expect to see near instant matchmaking. I do not. My guess is the underlying PSR number is used in the same fashion as elo. The only difference being the tier 1 players never see the those in the bottom two.

You are T1 right? Well you are fighting for T2/T3 that are all getting jammed into the other matches because there's higher population through the test of the tiers.

[Bilbo]

sycocys, on 30 December 2015 - 12:42 PM, said:

You are T1 right? Well you are fighting for T2/T3 that are all getting jammed into the other matches because there's higher population through the test of the tiers.

Nope. Still pushing through Tier 3. I get to play with/against everybody in the game.

[sycocys]

Playing as heavy?

My matches have been near instant since the voting change.

[Bilbo]

sycocys, on 30 December 2015 - 12:57 PM, said:

Playing as heavy?

My matches have been near instant since the voting change.

Assault

[AlphaToaster]

Are a lot of your games losing to non-combat victory conditions? Caps and whatnot? That would skew your W/L a bit.

I'm sitting at 1.34 W/L, with K:D of only 1.24. I do not consider myself one of the chosen few, so I don't try to be "that guy" every game. I'm less concerned about my score and more concerned about if we win or not in the moment. I'm also not above taking a technical win if my team is hopelessly down. I don't have any problems winning with a map victory condition. Do you have a "I will not cap win" rule?

[Bishop Steiner]

Tarogato, on 30 December 2015 - 11:20 AM, said:

Your win/loss ratio should correlate roughly to your kill/death ratio. At least, according to my research. So if you're bad at the game, you'll lose more. If you're good at the game, you'll win more. If you're good at the game and losing more, then you are a statistical anomaly and can take comfort in the knowledge that it will come to pass and you'll be out of the rut eventually.

this for Solo, Groups or in general?

I find group play tend to either artificially inflate (for larger competitive groups) or deflate (casual groups, 2 mans, etc) W/L substantially. Whereas a solo player, no matter how good, can only carry so hard, for so long.

For instance, I'm a solo player with a global KDr of 1.81 and a W/L of 1.18

[Revis Volek]

Bishop Steiner, on 30 December 2015 - 01:06 PM, said:

this for Solo, Groups or in general?

I find group play tend to either artificially inflate (for larger competitive groups) or deflate (casual groups, 2 mans, etc) W/L substantially. Whereas a solo player, no matter how good, can only carry so hard, for so long.

For instance, I'm a solo player with a global KDr of 1.81 and a W/L of 1.18

The only artificially inflated stat i see inflated at all are the w/l when running with groups. Its actually the opposite when it comes to DMG and KILLS when you run with a group of very competent mechwarriors. You have to work very hard most the time to get kills and dmg in before mechs die due to focus fire.

When i solo play, i have no issue getting high score and kills 800, 900, 1000 dmg and 1-4 kills is not unheard of. But in group queue im lucky to dish out 250-350 dmg and steal a kill because we are all on the same level and doing the same thing.

So again, the only thing i see that truly can get inflated artificially is your w/l, unless you had a team feeding you kills and dmg. I agree...

Edited by DarthRevis, 30 December 2015 - 05:05 PM.

[Tarogato]

Bishop Steiner, on 30 December 2015 - 01:06 PM, said:

this for Solo, Groups or in general?

I find group play tend to either artificially inflate (for larger competitive groups) or deflate (casual groups, 2 mans, etc) W/L substantially. Whereas a solo player, no matter how good, can only carry so hard, for so long.

For instance, I'm a solo player with a global KDr of 1.81 and a W/L of 1.18

Where were you guys when I was collecting this stuff a few months ago? hehe.

About 76 of the players reported being mostly solo players and only 5 players reported that they play mostly group queue. The rest didn't specify at all. Perhaps we can assume that solo players outnumber group players 15-to-1 in this sample, but take it with a very very large grain of salt.

[sycocys]

Bilbo, on 30 December 2015 - 01:03 PM, said:

Assault

Weird that's all I've been dropping and it's been instant.

[Zookeeper Dan]

Tarogato, on 30 December 2015 - 11:20 AM, said:

Your win/loss ratio should correlate roughly to your kill/death ratio. At least, according to my research. So if you're bad at the game, you'll lose more. If you're good at the game, you'll win more. If you're good at the game and losing more, then you are a statistical anomaly and can take comfort in the knowledge that it will come to pass and you'll be out of the rut eventually.

Cool data, but just eyeballing that line it looks off. It seems like it should be steeper.

[Bilbo]

Tarogato, on 30 December 2015 - 01:16 PM, said:

Where were you guys when I was collecting this stuff a few months ago? hehe.

About 76 of the players reported being mostly solo players and only 5 players reported that they play mostly group queue. The rest didn't specify at all. Perhaps we can assume that solo players outnumber group players 15-to-1 in this sample, but take it with a very very large grain of salt.

Bishop is in your list of people who's data you have and I'd have sworn I posted in that thread. At any rate a .92 k\d along side a 1.1 w/l would put me well off your line.

[EgoSlayer]

Bilbo, on 30 December 2015 - 12:33 PM, said:

Again, if this were the case I'd expect to see near instant matchmaking. I do not. My guess is the underlying PSR number is used in the same fashion as elo. The only difference being the tier 1 players never see the those in the bottom two.

Solo queue still has the 4x3 match maker conditions in addition to tier ranking so you still need multiple conditions to create a match. And if you don't see a 4x3 match that is the answer to why the match took longer to create - the release valves came off to allow a broader scope of mechs into the match. That happens first, before skill rankings come off. And according to the PSR release notes there isn't a complete release valve on PSR, the no Tier 1 vs Tiers 4/5 isn't ever supposed to happen.

[jss78]

I have a feeling that over the long haul, there's a tendency for the W/L to be somewhat greater than 1. There are two factors at play here.

First, most players will find themselves rising in PSR as they play. They start at a low tier, and gradually rise towards higher tiers. This means that all the way through this rise, they're playing at a somewhat "too low tier" for their skill level. Thus they have the opportunity to carry in their games -- maybe not much, but at least a little -- and they thus maintain a W/L of >1 through this process. At some point the player will finally reach his "correct" tier and start to win about 50% of games, but all those games racked up during the PSR rise will remain in the stats, and the W/L will remain at >1.

Second, as we all know, sometimes the MM puts you in a game where most other players are not from your tier. They might be worse than you, or better than you. However, as you get an experienced and get to a high tier, the MM can no longer use you as filler in games above your level, as such games don't exist: you'll only be facing players who are either at your level, or worse than you. So again, there's a tendency for you to get games where you can carry, and still maintain a W/L of >1.

I don't think MM quality should affect long-term W/L. Even if the MM sometimes makes lopsided games, over the long haul you should find yourself equally on both sides in those games. I don't think it credible that the MM would systematically work against or in favour of a specific player.

Edited by jss78, 30 December 2015 - 01:52 PM.

[Deathlike]

RNGesus when it comes to the MM will screw you over at times.

If you are actually "good", you'll generally win more than you lose and "feel" like you have more control over the situation than others.

[Bishop Steiner]

DarthRevis, on 30 December 2015 - 01:11 PM, said:

The only artificially inflated stat i see are the w/l when running with groups. Its actually the opposite when it comes to DMG and KILLS when you run with a group of very competent mechwarriors. You have to work very hard most the time to get kills and dmg in before mechs die due to focus fire.

When i solo play, i have no issue getting high score and kills 800, 900, 1000 dmg and 1-4 kills is not unheard of. But in group queue im lucky to dish out 250-350 dmg and steal a kill because we are all on the same level and doing the same thing.

So again, the only thing i see that truly can get inflated artificially is your w/l, unless you had a team feeding you kills and dmg.

I believe W/L was the stat we were talking about?

[Bilbo]

EgoSlayer, on 30 December 2015 - 01:45 PM, said:

Solo queue still has the 4x3 match maker conditions in addition to tier ranking so you still need multiple conditions to create a match. And if you don't see a 4x3 match that is the answer to why the match took longer to create - the release valves came off to allow a broader scope of mechs into the match. That happens first, before skill rankings come off. And according to the PSR release notes there isn't a complete release valve on PSR, the no Tier 1 vs Tiers 4/5 isn't ever supposed to happen.

I understand how it works. The only time I see unbalanced (by weight class) matches is when valves are released due to the introduction of new mechs.

[Dread Render]

it seems to me there may be a flaw in using KDR for anything.
Large slower mechs do lots of damage but may get fewer kills because they cant chase down a faster mech.
Faster mechs may tend to get more kills cause they can quickly take advantage of a damaged mech by getting to them fast and finishing them off.
If you are going to give weight to KDR it seems you also should take into account the mech that did the most.damage.

[EgoSlayer]

Appogee, on 30 December 2015 - 11:24 AM, said:

But is the matchmaker trying to create equal teams?

Or is it just collecting Tier 1, 2, and 3 players together, putting them in a team, and no effort made to average the skill level?

Incidentally, there's no way some of the players I'm getting dropped with are in Tier 3. I'd be surprised if they're even in Tier 4.

(But that still doesn't explain the seemingly uneven distribution of inexperienced players.)


My personal k/d is 1.65. I tend to get at least one kill per match, even though I'm presently levelling Mechs.

But like I said, doing 400+ damage and multiple kills fairly consistently is still leading to consistent team losses, for some reason.

The old match maker used a relative skill ranking (CN: It's a fixed size pool of skill ranking points and If you win you take ranking away from your opponent) system elo, to create what it thought were equal teams. Over time a working relative skill based ranking system will create conditions for eqaul skill matches meaning that you have an equal chance of winning or losing. Thus over time your W/L ratio will be close to 1.

PSR is not a relative skill based system, it's an experience based system. There is no fixed pool of points, your skill isn't compared to your opponents to determine points gained or lost, and you can sill go up on a loss. In fact this system rewards playing against lower tier opponents because the PSR gain is going to be larger when 'seal clubbing' than in a skill based system that offers little or no reward for playing against people in lower rankings.

Much like having two people who have played Baseball for the same amount of time (experience) they will have different levels of ability (Skill). So having twelve people of all the roughly same experience/Tier doesn't equal to them all having the same level of skill. Likewise for your opponents.

In short - no the current match maker won't result in a W/L ratio near 1 because it isn't using relative skills for the match making. My W:L rate has gone up 15% since PSR was implemented.