Forums / Community / Matchmaking Feedback & Discussion

[Locked] Matchmaking Feedback Update – May 21

OP ZaedynFel

  1. 1
  2. ...
  3. 38
  4. 39
  5. ...
  6. 40
eLantern wrote:
ZaedynFel wrote:
@Josh:

f077c259-eafd-4f4f-af95-072b8489e91c
242678af-abe9-4ef3-b005-d0b080b9af84

What were the odds of winning these games? (And were they really Onyxes?)
1) 75/25 for the winner. I see one Onyx pre-match, but the top two in kills after the match were both Onyx.
2) Winner was favored at 58% before the quitter, 94% after. Quitting reduces chances both because you have less players, but also because it serves as a signal that the quitting player was not playing as well as usual, which hurts their team. I see one Onyx on the losing team pre-match, who dropped to Diamond, and a Diamond who became Onyx. So really only one Onyx player in that match.
1. Out of curiousity, does the x/y predictions directly correlate to expected score? I.e. 75/25 equals predicted outcome of 50 to 17? And the 58% equate to 50 to 36?
2. I quit the second game when it was something like 35 to 9, and yet only lost 1 CSR. Bug? Or intentional?
3. Is the population skill cap limit (1.7 for Slayer, iirc) for the matchmaker on a per player basis, or per team? I.e. if I (as a roughly 1.0 player) were to team up with a couple 2.0 players, and another 1.0 player, would our average MMR be 1.5 [(2+2+1+1)/4] or 1.35 [(1.7+1.7+1+1)/4]?
1. Josh has mentioned previously that TS2 doesn't directly factor match spread, only wins and losses. It uses the team MMRs to determine the likely outcome percentage or ratio, but the individual expectation variances can indirectly equate to a score spread for slayer modes (only).

2. If you lost only 1 CSR it was because your MMR (at the time of quitting) was above your CSR. Also, since the odds shifted so heavily in favor of the opposing team the MMR impact to you likely became minimized so significantly that your MMR barely altered; in fact, if you managed to perform better than the changing circumstances (since individual performance is measured on a per minute level) you might have even been able to gain a minor amount of MMR prior to quitting assuming others on the team took the brunt of the negative impact for losing, but the action of actually quitting the match I'm sure impacted your MMR in a negative manner. The first quitter will always take the most significant impact in both MMR and CSR (-30).

3. That's a question Josh is more equipped to can handle.
1. A 50/50 game in which both teams are equally likely to win, I would expect to be a close 50 to 49 game (Not 50/50 just due to timing issues). I know individual variances are going to skew the actual score away from the predicted outcome. Also, there is an individual "kills per minute" variable...

2. I was the first and only to quit, so theoretically, I should have lost 30 CSR (I didn't).
eLantern wrote:
ZaedynFel wrote:
As for inflation, we'll keep using the current +15/-15, +1/-1 system to prevent inflation. We've been running simulations on alternative methods that are more forgiving in giving out points, but they all lead to unacceptable levels of inflation in the cases where win% gets above certain points. Win% can get high, of course, because of matchmaking deficiencies. But we'd like to keep Ranks correct when that happens.

So next season players will start about 200 below their target MMR and go +15/-1 until they get there. This will probably take around 20-60 games depending on win% and just randomness.
Just curious if you've tried running a simulation on a system that wouldn't award fixed amounts or simply 2 fixed amounts (+/- 1 or +/- 15)?

Wouldn't a system that could award a variable amount be best (at least within a limit)? If a person's CSR was above or below their MMR it could still reward +/-1 (per a win or a loss) in order to help prevent inflation or deflation, but it could also accurately award a variable amount of points up to a set limit (say 30?) that way players could get a better indication of their performance adjustments (via MMR) in-between matches.

Win, but barely outperform expectations and maybe you see your CSR go +7 in accordance with your MMR.

Win and perform above expectations (both team and individual) and you max out at +30 because your MMR jumped up more than that.

Win and over come the odds stacked against your team, but perform somewhat in line with expectations perhaps you only see a +19 CSR gain which would be less than some others on your team.

Similar situations with losses that could see you drop any variable amount in accordance with your MMR adjustment, but it would still be limited to a maximum drop of -30.

The +/- 30 would also mean that players would see their CSR converge to their MMR more quickly after a season reset (less grinding -- or is this what you want?) and the variable adjustments would mean that they could see the more minute changes in-between matches once they've converged their CSR to their MMR. That detail could also shed a little more light on the value of the expectations the system had for them per the match-up.

Another thought is to award several fixed set amounts between say a +/- 30 (or whatever limit you choose) depending on the MMR adjustment (assuming your CSR and MMR were converged).
+/- 1
+/- 5
+/- 10
+/- 15
+/- 20
+/- 25
+/- 30
I've toyed with this a little, but I didn't use +/- 1, I used +/- 5 whenever it would go the wrong direction, and the old update otherwise (1-30). This still inflated. But I could try +/- 1 instead.

I'm not sure I see a huge advantage over just the +15 though. I guess it could give a more intuitive number compared to your opponents.
Chimera30 wrote:
ZaedynFel wrote:
See the post before this about Juror. It's a rare case and won't cause problems for the vast majority of players.

As for inflation, we'll keep using the current +15/-15, +1/-1 system to prevent inflation. We've been running simulations on alternative methods that are more forgiving in giving out points, but they all lead to unacceptable levels of inflation in the cases where win% gets above certain points. Win% can get high, of course, because of matchmaking deficiencies. But we'd like to keep Ranks correct when that happens.

So next season players will start about 200 below their target MMR and go +15/-1 until they get there. This will probably take around 20-60 games depending on win% and just randomness.
I see. So that means some players will continue to see the +1 CSR for a win situation going on right now, if their CSR gets too high compared to their MMR? How far can CSR diverge from MMR before it's considered inflated/deflated? What are the benefits for continuing to use this "corrected CSR" ranking system over just showing MMR as the public rank? I ask because of how contentious the CSR correction fix has been.
I feel like if it's getting more than 2 tiers above where it should be (~100 MMR), that's getting inflated.

There are a couple of reasons for using CSR as is:
  1. H5's UI assumes CSR moves around within a certain range, and messing with that can have even more confusing consequences
  2. MMR can move more erratically than an expected +15, etc. It can move as fast as it feels it needs to. I've seen changes around 1000 in just one match. That would probably look crazy.
eLantern wrote:
ZaedynFel wrote:
@Josh:

f077c259-eafd-4f4f-af95-072b8489e91c
242678af-abe9-4ef3-b005-d0b080b9af84

What were the odds of winning these games? (And were they really Onyxes?)
1) 75/25 for the winner. I see one Onyx pre-match, but the top two in kills after the match were both Onyx.
2) Winner was favored at 58% before the quitter, 94% after. Quitting reduces chances both because you have less players, but also because it serves as a signal that the quitting player was not playing as well as usual, which hurts their team. I see one Onyx on the losing team pre-match, who dropped to Diamond, and a Diamond who became Onyx. So really only one Onyx player in that match.
1. Out of curiousity, does the x/y predictions directly correlate to expected score? I.e. 75/25 equals predicted outcome of 50 to 17? And the 58% equate to 50 to 36?
2. I quit the second game when it was something like 35 to 9, and yet only lost 1 CSR. Bug? Or intentional?
3. Is the population skill cap limit (1.7 for Slayer, iirc) for the matchmaker on a per player basis, or per team? I.e. if I (as a roughly 1.0 player) were to team up with a couple 2.0 players, and another 1.0 player, would our average MMR be 1.5 [(2+2+1+1)/4] or 1.35 [(1.7+1.7+1+1)/4]?
1. Josh has mentioned previously that TS2 doesn't directly factor match spread, only wins and losses. It uses the team MMRs to determine the likely outcome percentage or ratio, but the individual expectation variances can indirectly equate to a score spread for slayer modes (only).

2. If you lost only 1 CSR it was because your MMR (at the time of quitting) was above your CSR. Also, since the odds shifted so heavily in favor of the opposing team the MMR impact to you likely became minimized so significantly that your MMR barely altered; in fact, if you managed to perform better than the changing circumstances (since individual performance is measured on a per minute level) you might have even been able to gain a minor amount of MMR prior to quitting assuming others on the team took the brunt of the negative impact for losing, but the action of actually quitting the match I'm sure impacted your MMR in a negative manner. The first quitter will always take the most significant impact in both MMR and CSR (-30).

3. That's a question Josh is more equipped to can handle.
1. A 50/50 game in which both teams are equally likely to win, I would expect to be a close 50 to 49 game (Not 50/50 just due to timing issues). I know individual variances are going to skew the actual score away from the predicted outcome. Also, there is an individual "kills per minute" variable...

2. I was the first and only to quit, so theoretically, I should have lost 30 CSR (I didn't).
I have also had a couple of games where I quit first on my team, and did not get the -30 CSR penalty. I think this is an error in the system, unless they got rid of the -30 CSR penalty and I missed that info.
ZaedynFel wrote:
Actually, HCS has had the biggest increase vs. all playlists ever since the change, so we're not seeing a daily loss in players.
Wondering if that has to do with them adding all those new maps.
LUKEPOWA wrote:
ZaedynFel wrote:
Actually, HCS has had the biggest increase vs. all playlists ever since the change, so we're not seeing a daily loss in players.
Wondering if that has to do with them adding all those new maps.
I wonder if there are any plans to add those new maps to the slayer or team arena map rotations
  1. 1
  2. ...
  3. 38
  4. 39
  5. ...
  6. 40