26 replies to this topic

[Anomalocaris]

Nightbird, on 12 July 2020 - 10:29 AM, said:

The trolls are part of Cluster Fox's group, they're not gonna troll in their own member's thread lol

Oh man, now you went and spoiled it

[Cluster Fox]

I think we both agree the PSR system doesn't work.

We're proposing two different approaches to try to fix it.
Nightbird's system is totally worth a shot.
I'm trying to build on what Paul chose.

[Cluster Fox]

So, up for the challenge, I started simulating my own models. I'm using the same system as Nightbirds with one exception.

I still based the Win_Chance off the W-L stats. However, as opposed to Nightbird's model, I base the Stomp_Chance off the natural skill trendline seen in the player distribution. Player's ability to deal damage, break components and get kills is thus factored in Stomp chance. The ratio of W-L to avgMS is 1.58 times heavier for W-L, based on the standard deviation. Base values for both are still scaled from 200 to 1800.

There is a very strong correlation between stomp chance and win chance at 0.94362

Here is the baseline with the PSR reset. I ran 10000 matches but didn't update PSR at all. This serves as a reference for 100% blind MM where everyone is at PSR 2500 (day 0 of the reset)

The results for the baseline are very close to Nightbird's model - 25k games graph.

Processing the simulation takes a while so results will trickle in.

Edited by Cluster Fox, 18 July 2020 - 06:48 PM.

[Nightbird]

Looks good! I would start with simulating Jay Z's system first. Since we have live results already, it's possible to tune the parameters to match the current results, and use that as the basis of comparison for your suggestion.

[Cluster Fox]

It's about 2 days for 200k matches. I don't have specialized software so my methods are inefficient
I launched JayZ vanilla + CFox Stability for now, but I plan on running multiple scenarios.

[Nightbird]

Ah, do 25k matches. You'll find multiple tweak iterations needed, 2 days is just too much lol.

Also definitely start with Jay Z's system, otherwise you won't know whether your results are right or wrong.

[Cluster Fox]

Fair point, I was a big eager to see what Stability does.
I will try to closely match JayZ's actual results from the MM data first. In Tier movement and in WLR.

Edited by Cluster Fox, 19 July 2020 - 08:45 AM.