283 replies to this topic

[GweNTLeR]

Nightbird, on 29 June 2020 - 07:02 AM, said:

The graphs show out to 200K games. May I ask what the comparison to is, when you say the convergence is slow?

Well, you mentioned that your psr/wlr is calculated around 50 latest games. But well, I think that there will be an overregulation (no idea how big it would be however) which would require several hundred games to finish transient, during which stomps would occur often.
To make things worse, you eiter win or lose, so your PSR(WLR) would be quite unsteady and might change very rapidly.
So basically, if we are speaking about 60 games per month, 60 games sound ok, while 300 sound not so okay. Waiting for several more month is not really an option for MWO.
I mean, it is okay in general but I'm not sure if it would be so good for an individual.
So could you please make PSR(WLR)/games player by individual graph?

Edited by GweNTLeR, 29 June 2020 - 07:20 AM.

[Nightbird]

The simulation says that for the first month, WLR does better than the Community Proposal, and then it does even better with more time.

[Ignatius Audene]

Tldr. I just see your result and see all have w/l 1.0 over time. This is baaaaaad. Since mm will never be perfect (low population) the only way for a mm to achive w/l 1.0 for a good player is to give u all the potatoes in the world to carry until u lost enough.

And don't expect miracles. A mm that can only draw from 50 players with biiiiiig difference in skill will always result in suboptimal results.

Edited by Ignatius Audene, 29 June 2020 - 07:25 AM.

[Nightbird]

Ignatius Audene, on 29 June 2020 - 07:22 AM, said:

Tldr. I just see your result and see all have w/l 1.0 over time. This is baaaaaad. Since mm will never be perfect (low population) the only way for a mm to achive w/l 1.0 for a good player is to give u all the potatoes in the world to carry until u lost enough.

Nope sorry, how it works is by having newbies play newbies when possible, good players play good players when possible, and if a mix has to happen, make teams with 50/50 chances of winning.

Ignatius Audene, on 29 June 2020 - 07:22 AM, said:

And don't expect miracles. A mm that can only draw from 50 players with biiiiiig difference in skill will always result in suboptimal results.

That's what the simulation does (there's even a screenshot with players from PSR=5000 to 0), yet it succeeds. I guess I'm a miracle worker?

[GweNTLeR]

Nightbird, on 29 June 2020 - 07:20 AM, said:

The simulation says that for the first month, WLR does better than the Community Proposal, and then it does even better with more time.

Well, it surely does according to your summary graphs. But could you please show how it will work for an individual over games played? You cant run away from overregulation, you know...

[Ignatius Audene]

with 50/50 chances of winning.

Aka throw all the potatoes on u to carry.

[Nightbird]

GweNTLeR, on 29 June 2020 - 07:34 AM, said:

Well, it surely does according to your summary graphs. But could you please show how it will work for an individual over games played? You cant run away from overregulation, you know...

I'm not sure what overregulation means, here's individual pilots with 1 line per pilot. It's hard to look at, which is why I did the bar graph

[Collector of Mechs]

Thanks for making me internet famous. LOL. Great write up. Pretty pictures.
I know I am not a very good player. I would however like to point out one thing that not one of these PSR proposals take into account. (not suggesting they should either) A player like me, who plays a lot of QP, alone, with unskilled, or low skilled mechs trying to skill them up. Heck, if I get matched with new players, so I can earn more skill points faster and get away from the grind. Great! Most of the assumptions appear to be, everyone has been playing forever, and everyone has every mech fully skilled. I am probably the minority. Bought into game later than most, bought a bunch of mechs. Still adding skills to most. Usually, once I skill the mech, I move on to another. I don't know why. OCD I guess. Have a great day!

Edited by TheBones672, 29 June 2020 - 08:56 AM.

[GweNTLeR]

Nightbird, on 29 June 2020 - 08:02 AM, said:

I'm not sure what overregulation means, here's individual pilots with 1 line per pilot. It's hard to look at, which is why I did the bar graph

I meant overshoot, sorry, google translator
That's not really the graph I wanted to see, but nevertheless (would like to see like 20-50 players with the highest amount of games (or even some mediocre individual) and not the summary amount of games but amount games of played by those players on horisontal axis, with more than 8 slices).
The graph shows pretty good convergence after ~100k matches (that's like, 4 month?). Any chance you make the same graph with Jay Z proposal? Just to compare.

Edited by GweNTLeR, 29 June 2020 - 09:18 AM.

[Little Monk]

Where did you get their PSR data? I am pretty sure Jarl's doesn't have that information.

[Little Monk]

Additionally you fail to disclose how you are coming up with any of your calculations. Is a win simply the team with the higher match score? In which case this is a failed model since it is somewhat frequent to see a team with a higher match score losing.

How do you calculate % stomps? How do you calculate win chance? There are simply too many variable for which you are making very liberal assumptions.

[Nightbird]

GweNTLeR, on 29 June 2020 - 09:14 AM, said:

I meant overshoot, sorry, google translator
That's not really the graph I wanted to see, but nevertheless (would like to see like 20-50 players with the highest amount of games (or even some mediocre individual) and not the summary amount of games but amount games of played by those players on horisontal axis, with more than 8 slices).
The graph shows pretty good convergence after ~100k matches (that's like, 4 month?). Any chance you make the same graph with Jay Z proposal? Just to compare.

Here you go. I kept the same 50 players with the most games for both JZ and my method. Keep in mind they look "nicer" than the average pilot because of the number of games. The data snap shots are stored at every 25k games in the simulation so each pilot has 8 dots during the 8 snapshots. Note the Y axis tick marks.

Edited by Nightbird, 29 June 2020 - 11:33 AM.

[Nightbird]

Little Monk, on 29 June 2020 - 09:54 AM, said:

Where did you get their PSR data? I am pretty sure Jarl's doesn't have that information.

I started everyone at 2500 and let Jay Z's formula do the rest. That's what the PSR reset does right?

Little Monk, on 29 June 2020 - 10:07 AM, said:

Additionally you fail to disclose how you are coming up with any of your calculations. Is a win simply the team with the higher match score? In which case this is a failed model since it is somewhat frequent to see a team with a higher match score losing.

How do you calculate % stomps? How do you calculate win chance? There are simply too many variable for which you are making very liberal assumptions.

I linked my method in the beginning of the post. I chose not to repeat stuff about cumulative density functions and randomization methods in this thread.

[Nightbird]

Question from reddit: https://www.reddit.c...cy_with_graphs/

Quote

Thank you for your work! I can only hope the community pushes for a WLR-oriented PSR system like yours (with revisions if needed). I was unaware that JayZ's proposal would bring about that broad a curve for percentage of players vs. WLR among the player base, so I appreciate you running a simulation for it. I'm curious, though, how does the current percentage of players vs. WLR curve compare to that produced by JayZ's proposal? I'm hopeful it's a statistically significant increase for JayZ, since it seems we'll be running with their system for a while.

I kept the 10k players from season 47 with the most games just like my simulation, to reduce the outliers.



I think Jay Z's proposal is the same as the current MM, it will just take 2-3 months to get there. Maybe a bit longer because I reduced the pilot roster for my simulation.

Edited by Nightbird, 29 June 2020 - 12:01 PM.

[GweNTLeR]

Nightbird, on 29 June 2020 - 11:23 AM, said:

Here you go. I kept the same 50 players with the most games for both JZ and my method. Keep in mind they look "nicer" than the average pilot because of the number of games. The data snap shots are stored at every 25k games in the simulation so each pilot has 8 dots during the 8 snapshots. Note the Y axis tick marks.

Now that's looking awesome, thanks!
Your proposal is surely more unstable compared to Jay Z (I'm pretty sure it is overshoot in the beginning), but overall - actually looking great!
On the second thought - errr it takes 2k matches individually to get somewhat stable? Sounds like too much for me.

Edited by GweNTLeR, 29 June 2020 - 12:47 PM.

[Nightbird]

GweNTLeR, on 29 June 2020 - 12:39 PM, said:

On the second thought - errr it takes 2k matches individually to get somewhat stable? Sounds like too much for me.

No, it's just 60 games. Stability depends on the rest of the player base getting games in. One pilot putting up more games means nothing. Also note that the Y axis is 3x wider in Jay Z's plot, so it's not more stable just vertically compressed.

We're going to be seeing Jay Z's system regardless though, so there's no choice to be had here.

Edited by Nightbird, 29 June 2020 - 01:00 PM.

[MadcatX]

A shame your system wasn't even given consideration by PGI, at least that's the impression I got. One thing I'm curious about is, why are you still here? I do not mean that in a negative way. I thought I had heard that your were an inactive player and when I read your first rebuttal when they announced what they were going to do (love the whole "ok, try this with MWO2 comment"), took that as being your "meh, tried my hardest, time to move on" statement.

[Nightbird]

MadcatX, on 29 June 2020 - 04:22 PM, said:

A shame your system wasn't even given consideration by PGI, at least that's the impression I got. One thing I'm curious about is, why are you still here? I do not mean that in a negative way. I thought I had heard that your were an inactive player and when I read your first rebuttal when they announced what they were going to do (love the whole "ok, try this with MWO2 comment"), took that as being your "meh, tried my hardest, time to move on" statement.

Fair question, I reinstalled because a few buddies asked me to. I'm waiting for another game (SC) so it doesn't hurt to have another game to play while waiting for the alpha patches.

The biggest WHY is because I enjoy the MW franchise. The dumb ****** system I suggested, if implemented at the beginning of MWO, would have increased total revenue by 50% or more, 100% is not impossible. If MWO has earned 50 million $ to date (over 8 years), I'm talking about an idea that would have generated an additional 25-50 million $. This means promised features might have had the budget to be implemented. I'm hoping PGI recognizes the importance of optimizing the user experience, but like you I feel like they don't get it.

[Kamikaze Viking]

I just hope they look at this and consider it and storing more relevant data for MWO2's MM

[ccrider]

Read whole thread. Looked at the compare/contrast and data Nightbird put up. Impressed. Math don't lie. Nightbird's system is better. Unfortunately the one real takeaway I got from all this was, hahaha, MWO2. Yeah, that'll definitely happen. I envy the true believers, but after all these years I'm confident this is all this game will ever be.