713 replies to this topic

[The Teddy Bear]

Andrzej Lechrenski, on 10 June 2020 - 04:05 PM, said:

No, I understood. But facts do not matter how much a player feels he 'should have' contributed. He either contributed or he didn't. If he contributed, that will show up. If he didn't, that will to. End of story. If a self-described "good player" feels like their W/L ratio has gone down, then they are clearly not able to keep up with the current meta. But you haven't even been playing the game for the last two years, so you are wasting everybody's time here when you pretend to be speaking from experience and yet are not. Play this season, and then, when you have a stake in the game again, we can talk about this with a common frame of reference. That would be productive. This is not.

Sorry but I have done that. W/L is a **** load of it and you can continue arguing this. In the old solo que W/L would have been fine for me. Involving groups, and still arguing for W/L and I am sorry for you to destroying the game.

[Nightbird]

Socal Bronco, on 10 June 2020 - 03:53 PM, said:

This thread is supposed to be proposals, not arguing over whether AMS should be in match scores. Most of the proposals are pointless because they don't explain how the data will be used to assign teirs. I've only seen 2 proposals that explain how Tiers would be decided. Nightbirds that would use WL ratio, however he doesn't explain where he would draw the lines for the Tiers and whether they would be permanently set or whether they would change over time.

The other proposal was mine where I laid out that what ever measure you use would be broken down by % of player base to keep T1 at 5 % of players.
We also don't know how they do that now.

Check the link, there is no hard tier breaks, just sort on WLR easy peasy

Did you explain how the MS solution can create better matches than the current MM? My second link proved that it can't.

Andrzej Lechrenski, on 10 June 2020 - 04:00 PM, said:

If a "great player" who can only bother to play the game 30 matches a month doesn't get the idolization he thinks he deserves... I'm not going to shed a single tear. Not one.

5 games a day is not 30 a month? And it's relative, someone playing 500 games a month will still be eclipsed by someone playing 3000 a month, do you really want the #games played to have that much influence in a measurement of skill?

The Teddy Bear, on 10 June 2020 - 04:11 PM, said:

Sorry but I have done that. W/L is a **** load of it and you can continue arguing this. In the old solo que W/L would have been fine for me. Involving groups, and still arguing for W/L and I am sorry for you to destroying the game.

On the basis that you said so therefore it is true?

Edited by Nightbird, 10 June 2020 - 04:22 PM.

[The Teddy Bear]

Nightbird, on 10 June 2020 - 04:15 PM, said:

On the basis that you said so therefore it is true?

Nope, Just come from experience.

[Nightbird]

LordNothing, on 10 June 2020 - 04:04 PM, said:

400 is certainly an above average player according to jarls global stats. how do i know the wins and losses arent due to other players in that match? the 10 wlr would indicate to me that this player ended up in matches with similar or higher skilled players. the guy with 2 wlr probably lost because he had to carry taters.

my concern is the 50 ms scrub getting credited for a win and gaining psr they didnt really deserve and which is going to lead to worse games in the future.

i also dont like that we are nitpicking the details when the important thing is the zero sum, which will allow movement to the correct tier for the skill, no matter how skill is determined.

The other players on your team will average out in skill over time. The only consistent performer is you. Over the long term, your contribution is your WL.

Under the WLR system, if the system rates you too high, you can't carry your weight and you drop down. If it rates you too low, you win more and float up. The MS system doesn't do that, since win or lose you will earn or lose points on average, it doesn't pin you to a certain value or anything else. At any time, the ranking of all players is almost purely random, as much based on # games played as skill.

Edited by Nightbird, 10 June 2020 - 04:23 PM.

[Nightbird]

The Teddy Bear, on 10 June 2020 - 04:18 PM, said:

Nope, Just come from experience.

Ok, anecdotal then, versus an analysis from a database including the performance of all players over multiple seasons saying WLR is by far the best.

Edited by Nightbird, 10 June 2020 - 04:20 PM.

[Andrzej Lechrenski]

Nightbird, on 10 June 2020 - 04:15 PM, said:

5 games a day is not 30 a month? And it's relative, someone playing 500 games a month will still be eclipsed by someone playing 3000 a month, do you really want the #games played to have that much influence in a measurement of skill?

You got me. I let my annoyance at farmers get the better of me.

[Nightbird]

Andrzej Lechrenski, on 10 June 2020 - 04:25 PM, said:

You got me. I let my annoyance at farmers get the better of me.

My proposal makes farming much much harder. Again, proof in the first link.

That I am a farmer is not by choice, but by decree. By PGI. Through how they create their game systems.



Even for FP, I long argued for tonnage to be set after analyzing the WLR of both teams, to make the results as close a possible. Pop's destiny is death by negligence

Edited by Nightbird, 10 June 2020 - 04:33 PM.

[The Teddy Bear]

Nightbird, on 10 June 2020 - 04:20 PM, said:

Ok, anecdotal then, versus an analysis from a database including the performance of all players over multiple seasons saying WLR is by far the best.

You did not even understand the basic did you. A "good" group player will get themselves into tier 1 in a second as long as they surround themselves with "decent" players. A "good" solo player will not get there for quite a long time, if w/l are on line, as they cannot influence the games as them could do in the old solo que. The worst is that a "bad" solo player will most likely have increased their w/l, due to their abilltiy to loose matches, are not impacting a match as they did earlier.....

Edited by The Teddy Bear, 10 June 2020 - 04:35 PM.

[Nightbird]

The Teddy Bear, on 10 June 2020 - 04:32 PM, said:

You did not even understand the basic did you. A "good" group player will get themselves into tier 1 in a second as long as they surround themselves with "decent" players. A "good" solo player will not get there for quite a long time, if w/l are on line, as they cannot influence the games as them could do in the old solo que. The worst is that a "bad" solo player will most likely have increased their w/l due to their abilltiy to loose matches, are not impacting a match as they did earlier.....

I just read that a good group player will win more than a good solo player, even when both are the same skill. I agree with this.

Do you think the match maker should have the same expectation from both players when assigning them to teams, and ignore the fact that one plays with a group? The MS system will do that, and continue to under-estimate groups. If that is your goal then keep trucking.

Edited by Nightbird, 10 June 2020 - 04:36 PM.

[Gagis]

Hey, I was trying to propose a system here, but instead got drowned out by pointless bickering.

Get a room.

[The Teddy Bear]

Nightbird, on 10 June 2020 - 04:36 PM, said:

I just read that a good group player will win more than a good solo player, even when both are the same skill. I agree with this.

Do you think the match maker should have the same expectation from both players when assigning them to teams, and ignore the fact that one plays with a group? The MS system will do that, and continue to under-estimate groups. If that is your goal then keep trucking.

This was my original post:

Sorry, but I dont feel like reading through 15 pages at all. Can someone pls - summarize up the best suggestion for PSR changes? We can hopefullly line up from that point. Delivering PSR alternatives and what´s producing match score in the same thread seems like "bring back the kindergarten" to me.

And I dont remember seeing any answers apart from yours, which i did not like at all.

Edited by The Teddy Bear, 10 June 2020 - 04:44 PM.

[Nightbird]

The Teddy Bear, on 10 June 2020 - 04:42 PM, said:

This was my original post:

Sorry, but I dont feel like reading through 15 pages at all. Can someone pls - summarize up the best suggestion for PSR changes? We can hopefullly line up from that point. Delivering PSR alternatives and what´s producing match score in the same thread seems like "bring back the kindergarten" to me.

And I dont remember seeing any answers apart from yours, which i did not like at all.

Then lay-off the criticisms without any evidence to back it up, it's like you're trying to get me to reply.

Gagis, on 10 June 2020 - 04:41 PM, said:

Hey, I was trying to propose a system here, but instead got drowned out by pointless bickering.

Get a room.

Sorry man, feel free to repost.

[Andrzej Lechrenski]

Nightbird, on 10 June 2020 - 04:28 PM, said:

My proposal makes farming much much harder. Again, proof in the first link.

That I am a farmer is not by choice, but by decree. By PGI. Through how they create their game systems.



Even for FP, I long argued for tonnage to be set after analyzing the WLR of both teams, to make the results as close a possible. Pop's destiny is death by negligence

Surely you have noticed that I've been agreeing with at least the principle of your proposal (which, obviously, yes I have read).

[LordNothing]

Socal Bronco, on 10 June 2020 - 03:53 PM, said:

This thread is supposed to be proposals, not arguing over whether AMS should be in match scores. Most of the proposals are pointless because they don't explain how the data will be used to assign teirs. I've only seen 2 proposals that explain how Tiers would be decided. Nightbirds that would use WL ratio, however he doesn't explain where he would draw the lines for the Tiers and whether they would be permanently set or whether they would change over time.

The other proposal was mine where I laid out that what ever measure you use would be broken down by % of player base to keep T1 at 5 % of players.
We also don't know how they do that now.

using an ms based system would probably also come with a major tweaking of the individual payouts towards match score. this is a good thing because some of those payouts need to be tweaked anyway, climbing psr because you run a lot of ams might not be too good, especially when you find yourselves with a bunch of meta players on the other team who dont use lerms. but thats easily fixed by tweaking the ams payout. ms is good because it turns as many win conductive behaviors into a single value. and it also compares

for implementation take the sum ms of your last 100 games. if you have fewer then 100 compute a fudge factor 100/games played (though ommiting entirely it might be good for new players as they would always be at the bottom of the curve until they get their 100 games in, thought it might be bad if you take alt accounts into consideration), divided by 100.

then you are placed on a bell curve using standard deviations which determine your tier. i think this is better than using the flat 20% population blocks as the further from average you get the less players there are. the center 20% is going to have a lot less variation than the next 20% because thats where the curve gets steep. t2 and 4 would have way too much variation in skill and things flatten out again at the extremes. id go with this method for tier placement no matter what is measured, if its w/l you could use that directly with no further modification as the values average over time.

my major concern with w/l is my massive backlog with 26k games, and i screw around a lot and level trash mechs (i mastered all pre skill tree mechs that you could get for cbills, and boy was there some trash in there) so not all those games are played to win, especially ones spent grinding events. but if you stick me in t4 im going to be mopping the floor repeatedly and it would take about another 10k games to get properly placed. so if this is used you must limit its scope enough where only a portion of the most recent backlog is used. like your last 100 games.

Edited by LordNothing, 10 June 2020 - 05:10 PM.

[Nightbird]

Andrzej Lechrenski, on 10 June 2020 - 04:51 PM, said:

Surely you have noticed that I've been agreeing with at least the principle of your proposal (which, obviously, yes I have read).

Yep, thanks as well, it was just in case people misunderstood the farming..

[The Teddy Bear]

Nightbird, on 10 June 2020 - 04:49 PM, said:

Then lay-off the criticisms without any evidence to back it up, it's like you're trying to get me to reply.

You did just reply, didn´t you?

I just feel like there should still been some decent players in here, that will basically disregard the w/l, and go for a stat more like matchscore as it is right now with the combined que. And then, if we get the possibility, refine the balance after the outcome.

[Nightbird]

LordNothing, on 10 June 2020 - 04:52 PM, said:

for implementation take the sum ms of your last 100 games. if you have fewer then 100 compute a fudge factor 100/games played (though ommiting entirely it might be good for new players as they would always be at the bottom of the curve until they get their 100 games in, thought it might be bad if you take alt accounts into consideration), divided by 100.

then you are placed on a bell curve using standard deviations which determine your tier. i think this is better than using the flat 20% population blocks as the further from average you get the less players there are. the center 20% is going to have a lot less variation than the next 20% because thats where the curve gets steep. t2 and 4 would have way too much variation in skill and things flatten out again at the extremes.

This is better than most other MS proposals. When compared to WLR it is worse because past MS is weakly correlated to future wins whereas past WLR is strongly correlated. If implemented, this idea is the same as using avgMS (past 100 matches) and will result in about 20% the improvement that the same implementation using WLR would bring.

You can read the second link I posted with the graphs or consider this simple example.

Two players played 100 games and have 400avgMS, one has 10 WLR and the other has 2 WLR. You are to predict their WLR for their next 100 games. If you predict 6WLR for both players, that is the MS system. If you predict 10 WLR and 2 WLR, then that is the WLR system. This is because MS considers differences in WLR to be accidental, and that the true skill of the players are presented by their MS.

Edited by Nightbird, 10 June 2020 - 05:08 PM.

[Andrzej Lechrenski]

Gagis, on 10 June 2020 - 04:41 PM, said:

Hey, I was trying to propose a system here, but instead got drowned out by pointless bickering.

Get a room.

Jay Z hasn't actually proposed his system yet. At least, not in this thread. So you are way ahead of him, despite the bickering. However, if you could make a Google Doc spreadsheet or some graphs or something, that would be great.

[Andrzej Lechrenski]

LordNothing, on 10 June 2020 - 04:52 PM, said:

using an ms based system would probably also come with a major tweaking of the individual payouts towards match score. this is a good thing because some of those payouts need to be tweaked anyway, climbing psr because you run a lot of ams might not be too good, especially when you find yourselves with a bunch of meta players on the other team who dont use lerms. but thats easily fixed by tweaking the ams payout. ms is good because it turns as many win conductive behaviors into a single value. and it also compares

for implementation take the sum ms of your last 100 games. if you have fewer then 100 compute a fudge factor 100/games played (though ommiting entirely it might be good for new players as they would always be at the bottom of the curve until they get their 100 games in, thought it might be bad if you take alt accounts into consideration), divided by 100.

then you are placed on a bell curve using standard deviations which determine your tier. i think this is better than using the flat 20% population blocks as the further from average you get the less players there are. the center 20% is going to have a lot less variation than the next 20% because thats where the curve gets steep. t2 and 4 would have way too much variation in skill and things flatten out again at the extremes. id go with this method for tier placement no matter what is measured, if its w/l you could use that directly with no further modification as the values average over time.

my major concern with w/l is my massive backlog with 26k games, and i screw around a lot and level trash mechs (i mastered all pre skill tree mechs that you could get for cbills, and boy was there some trash in there) so not all those games are played to win, especially ones spent grinding events. but if you stick me in t4 im going to be mopping the floor repeatedly and it would take about another 10k games to get properly placed. so if this is used you must limit its scope enough where only a portion of the most recent backlog is used. like your last 100 games.

I don't know if I agree that is how standard deviations work. Using standard deviations also won't help very much with such a low population, as it essentially means that either the extremes will never get matches, or else the MM will be loosened so much that it was a wasted effort to begin with. Might as well just go with a percentage based system where top W/L is 99.9999+% and bottom is 0.0001% and just let them loosen valves from there.

Considering using only the most recent hundred or so matches is a fine idea, though.

[VerKa]

I'll probably get flamed for this. Just take your original patch you were going to implement on Tuesday and install it. Run for a week, tweak it . Run it another week , do the same. You have to start somewhere. Thanks.