51 replies to this topic

[adamts01]

MrMadguy, on 16 January 2018 - 12:55 AM, said:

You know, best proof, that you're wrong - is leaderboard itself. All, W/L based MM cares about - to make W/L as close to 1, as possible, cuz W/L - is only variable in equation. And the best player with W/L = 1 in our leaderboard has AvgMS = 451, while the worst one has 37... Yeah. And according to W/L based MM they all would have equal skills, lol.

You're overthinking things. Ideally everyone would be close to 1, regardless of their tier, that means fights are pretty balanced. If someone started doing great in tier 2 and started to get a great W/L then he could bump an under-performer out of tier 1. Once that tier 2 guy got to tier 1 he should see his W/L drop, hopefully to around 1. So a tier 3 player with a 1 W/L wouldn't be nearly as good as a tier 1 player with a .88 W/L. Does that make sense?

[BTGbullseye]

In a W/L only rating:

1. Get in a group, and do nothing at all, become T1 by sitting on your hands.
2. Get in a group, do excellent personally but your team still loses, become T5 despite being one of the better players.

W/L is the WORST single-point rating metric available in this game.

[adamts01]

BTGbullseye, on 16 January 2018 - 02:47 AM, said:

In a W/L only rating:

1. Get in a group, and do nothing at all, become T1 by sitting on your hands.
2. Get in a group, do excellent personally but your team still loses, become T5 despite being one of the better players.

W/L is the WORST single-point rating metric available in this game.

Obviously this is a qucik play thing, especially since group que doesn't really exist anymore.

[MrMadguy]

adamts01, on 16 January 2018 - 01:03 AM, said:

You're overthinking things. Ideally everyone would be close to 1, regardless of their tier, that means fights are pretty balanced. If someone started doing great in tier 2 and started to get a great W/L then he could bump an under-performer out of tier 1. Once that tier 2 guy got to tier 1 he should see his W/L drop, hopefully to around 1. So a tier 3 player with a 1 W/L wouldn't be nearly as good as a tier 1 player with a .88 W/L. Does that make sense?

I would make simulator to prove, that you're wrong, but I have no time for this and I also hate burden of proof shifting. You need to prove, that you're right - you make simulator, that will prove it. If you can't, then, yeah, you don't have anything else, than mere words and faulty assumptions. Till then we will consider, that you're wrong. And stop this ELO-trolling already.

Edited by MrMadguy, 16 January 2018 - 05:15 AM.

[BTGbullseye]

adamts01, on 16 January 2018 - 02:53 AM, said:

Obviously this is a qucik play thing, especially since group que doesn't really exist anymore.

The entire PSR "system" is QP exclusively... And the "group queue" is the same as PUG, you're just there with other people that are part of your team. If you go in a 12-man premade against PUGs, (entirely possible, and I've seen it at least once in the last 2 weeks) you can have one of the premade sit on their hands, and still win.

Edited by BTGbullseye, 16 January 2018 - 04:14 AM.

[arcana75]

BTGbullseye, on 16 January 2018 - 04:14 AM, said:

The entire PSR "system" is QP exclusively... And the "group queue" is the same as PUG, you're just there with other people that are part of your team. If you go in a 12-man premade against PUGs, (entirely possible, and I've seen it at least once in the last 2 weeks) you can have one of the premade sit on their hands, and still win.

Indeed. Of all the stats, match score is the one that most accurately represents a player's performance and thus is currently the best marker for player skill.

[adamts01]

MrMadguy, on 16 January 2018 - 04:06 AM, said:

I would make simulator to prove, that you're wrong,

I'd love to see it, but on top of my lack of skills on a computer I also have a lack of faith that anything can save this game. On a similar note, I just noticed this post. https://mwomercs.com...is-of-the-12-0/
The most telling of all stats determining a win was the average W/L of the teams, next being match score, followed by average kills per match. As someone interested in such things, you should check out the massive undertaking that player took on.

[sycocys]

I wish I wasn't laughing so hard that I forgot to take a screenshot.

24 damage, 9 assists (no idea how with 24 damage didn't see any lrms on our side) + a win = PSR up. You can be a complete potato playing a light mech with a controller and barely being able to remember which button fires and still get to T1 with this system because that single up is still more than multiple downs for doing similar output on losses.

[arcana75]

adamts01, on 17 January 2018 - 04:22 AM, said:

I'd love to see it, but on top of my lack of skills on a computer I also have a lack of faith that anything can save this game. On a similar note, I just noticed this post. https://mwomercs.com...is-of-the-12-0/
The most telling of all stats determining a win was the average W/L of the teams, next being match score, followed by average kills per match. As someone interested in such things, you should check out the massive undertaking that player took on.

Very good post by Tarogato. One thing though, even though his emphasis was on WL ratio being the closest correlator, MS was a VERY CLOSE 2nd. As well, I found it intriguing that he used MS as an example to illustrate a better MM algorithm instead of WL ratio.

Let me ask this. If you were in a group of friends doing solo QP, be it sync drop or group, if you had to pick JUST ONE, which stat do you think is positively influenced the most by being in a group, your WL ratio or your MS? Pick just one, even if it's damn close.
.
.
.
.
The opposite of what you picked should be one used for Tiering and matchmaking in solo QP.

Edited by arcana75, 17 January 2018 - 05:56 PM.

[sycocys]

arcana75, on 17 January 2018 - 05:56 PM, said:

Let me ask this. If you were in a group of friends doing solo QP, be it sync drop or group, if you had to pick JUST ONE, which stat do you think is positively influenced the most by being in a group, your WL ratio or your MS? Pick just one, even if it's damn close.
.

W/L goes up with groups.
MS goes down, unless you normally post really terrible scores.

That's just the nature of teamwork.

-- player skill ratings should be use secondary if even at all in solo QP. Solo's primary focus should be creating balanced matches at the base level.

[adamts01]

arcana75, on 17 January 2018 - 05:56 PM, said:

Very good post by Tarogato. One thing though, even though his emphasis was on WL ratio being the closest correlator, MS was a VERY CLOSE 2nd. As well, I found it intriguing that he used MS as an example to illustrate a better MM algorithm instead of WL ratio.

Let me ask this. If you were in a group of friends doing solo QP, be it sync drop or group, if you had to pick JUST ONE, which stat do you think is positively influenced the most by being in a group, your WL ratio or your MS? Pick just one, even if it's damn close.
.
.
The opposite of what you picked should be one used for Tiering and matchmaking in solo QP.

I think there is a place for match score, especially with a small sample size of games and because MS has gotten so much better over the years, but in the long run I still think W/L is more telling. As primarily a light player I can't tell you the number of times I've left a perfect position to help defend my assaults and finish the match with pathetic damage because of fighting off those enemy lights. Or running across the map to suicide in order to stop a base cap to give my heavies time to get there. Or held my fire so I could continue holding a lock for LRMs, ending with close to no damage. Or continued to drop call after a double gauss hit my cockpit before I ever fired a shot. Those aren't just group que sacrifices, so I don't agree with your logic that W/L is any more or less a factor depending on which que you're in. I also think his algorithm is solid using W/L in combination with everything else, and probably much better than pure W/L. I just see W/L as an incredibly simple method that PGI would have to try really hard to mess up, and I think it would be monumentally better than what we currently have, if combined with a system where you had to knock someone down if you wanted to move up.

[MrMadguy]

adamts01, on 17 January 2018 - 04:22 AM, said:

I'd love to see it, but on top of my lack of skills on a computer I also have a lack of faith that anything can save this game. On a similar note, I just noticed this post. https://mwomercs.com...is-of-the-12-0/
The most telling of all stats determining a win was the average W/L of the teams, next being match score, followed by average kills per match. As someone interested in such things, you should check out the massive undertaking that player took on.

While I've also referred to some stats from leaderboard, we should be very careful, when doing it, as we can't take MM itself out from this equation and match making works both ways - players' stats can affect matchmaking, but matchmaking can also affect players' stats. I.e. in this case we can't be sure, that there is no positive feedback in MM, that caused this correlation - we don't know for sure, what is cause and what is consequence in this case. This correlation can be caused by initial flaw in MM, that allowed some players to win more, than they should have been, and increased average W/L was just consequence of this initial flaw - not due to higher skill.

adamts01, on 17 January 2018 - 06:49 PM, said:

I think there is a place for match score, especially with a small sample size of games and because MS has gotten so much better over the years, but in the long run I still think W/L is more telling. As primarily a light player I can't tell you the number of times I've left a perfect position to help defend my assaults and finish the match with pathetic damage because of fighting off those enemy lights. Or running across the map to suicide in order to stop a base cap to give my heavies time to get there. Or held my fire so I could continue holding a lock for LRMs, ending with close to no damage. Or continued to drop call after a double gauss hit my cockpit before I ever fired a shot. Those aren't just group que sacrifices, so I don't agree with your logic that W/L is any more or less a factor depending on which que you're in. I also think his algorithm is solid using W/L in combination with everything else, and probably much better than pure W/L. I just see W/L as an incredibly simple method that PGI would have to try really hard to mess up, and I think it would be monumentally better than what we currently have, if combined with a system where you had to knock someone down if you wanted to move up.

You should clearly understand, that "scouting" Lights are rather exception from rule, than rule. This game is shooter and therefore it's about killing other 'Mechs and dealing damage. And PGI have actually realized, that QP can't provide right experience for such players, and implemented dedicated mode for such players - Scouting Mode.

Edited by MrMadguy, 17 January 2018 - 10:13 PM.