87 replies to this topic

[Nightbird]

To clarify, you shouldn't see a big spike in the final WLR graph if the win-rate brackets are so depressed in the center. Having a 45-55 center bracket value around 10% gave me Jay Z's win distribution.

Edited by Nightbird, 14 October 2020 - 07:02 AM.

[ShiverMeRivets]

Well I just crossed the threshold into tier 1.
The new system doesn’t work.

It seems like player tend to slowly drift to the tier-scale edges, either up or down.

(Played PXH-Roc a few games, that sent me back to tier 2 )

Edited by ShiverMeRivets, 14 October 2020 - 08:09 AM.

[Nightbird]

ShiverMeRivets, on 14 October 2020 - 08:07 AM, said:

Well I just crossed the threshold into tier 1.
The new system doesn’t work.

It seems like player tend to slowly drift to the tier-scale edges, either up or down.

(Played PXH-Roc a few games, that sent me back to tier 2 )

You would bounce between T3 and T2 under the WLR PSR I proposed.

(based on your Jarl stats)

[Cluster Fox]

Nightbird, on 14 October 2020 - 07:01 AM, said:

To clarify, you shouldn't see a big spike in the final WLR graph if the win-rate brackets are so depressed in the center. Having a 45-55 center bracket value around 10% gave me Jay Z's win distribution.

Disclaimer : This message answers to Nightbird and will be gibberish for a lot of people.
To explain a significant variation in my model and yours:
My MM simulation uses a flip-flop sorting mechanism to balance teams.
Selected 24,48,72 or 96 players are all ordered by PSR, then placed in teams of 12 like this:
1 - 2 - 2 - 1 - 1 - 2 - 2 - 1 - 1, etc.

So there is no actual effort by my simulated MM to equalize both team's total PSR. It's a dirty approximation but it's quick and still delivers. Maybe the MM we currently have behaves like this, maybe I'm way off, we just don't know.

Since this is an approximation, tweaking was required. I had to tighten the Win-Loss skill Delta compared to your model to arrive at the real seasons stats. This explains the difference in win chance distribution. With a wider Win-Loss Delta, I get nice bell curves like your model (very closely matched actually) but I'm straying further from actual seasonal data.

This method also has a bias towards Team 1 which is captured in my win-chance bar graph.

As a result of the method, my simulated MM can't balance extreme PSR spreads very well, hence the spikes.

Long story short, my win-chance can't be measured against your win-chance because your MM is more exact.
However, PSR systems can be compared relative to each other because they use the same data, MM and skill gap.

Edited by Cluster Fox, 14 October 2020 - 09:30 PM.

[Nightbird]

Ah I see, thanks for the response!

[MechWarrior414712]

I have come to beat the dead horse with my stick

Time to delete the noobs from tier 1-3 and send them to tier 5

[Capt Deadpool]

I O O percent KongLord, on 15 October 2020 - 12:19 PM, said:

I have come to beat the dead horse with my stick

Time to delete the noobs from tier 1-3 and send them to tier 5

The increased wait times would likely drive players away that have been sticking through everything so far.

edit: Well I guess moving ppl around between tiers probably wouldn't affect wait times because population is so low that tiers are being mashed together anyway. But I suppose the implication of your post was to eliminate a number of less skilled players from Tier 1, and then prevent MM from mashing tiers together... but that would lead to long wait times under the current system due to current game population, as I understand it. I hope PGI will reconsider some of the excellent work done by Nightbird and Clusterfox for MWO or MWO2 when it comes to implementing a better solution

With the wait times I had last night with just a two-man group, during peak North America hours no less, I suspect if it were much worse it would drive people away.

Edited by Capt Deadpool, 15 October 2020 - 11:18 PM.

[Cluster Fox]

If games drive potential new players away because they get punished for trying to play, we're not better off.