Wow, I love the new per-person charts. They make so much more sense, visually. Haven’t had a look at the MU chart overall yet, but wanted to call out just how much better the personal charts looked. Very cool!
One of my big takeaways from the character skill charts is that I should stop playing Pers in tournaments (or double down on her), because I’m pretty pants at the moment.
Oh, one change I would make is to show all of the individual charts on the same scale. Makes comparison easier.
Yeah, fair point. I’ll tweak it.
Skilled enough to die to your mono-throw Vendetta in tournament.
this one looks a lot closer to what I in my heart “know” to be correct. Biggest numbers off from what I “know” are quince-geiger zane-oni arg-lum BBB-Midori Midori-Troq. Basically I think there are several hard CPs that are only hard CPs if played in a very specific way that the model probaly doesn’t respect the top players win % as much. I’d like to see a +/- for each char (how far away from 400) these sums are and compare it to our chart to see.
Chart correctly pegs Oni-Arg as better than Oni-DeGrey, and that Rook-Arg is a bad MU. Very strong chart imo.
edit: Woah, it even puts it lower than Rook-BBB!
incorrectly has it for lum though. I’m convinced Degrey is better vs Oni and Arg is better vs Lum and feel pretty strongly about both
Fair enough. Lum has been a fairly unpopular character since I started tracking data, so I’m not surprised that an analysis based on that data isn’t as well refined as for other characters though.
Is there a chance of getting a person specific version for me?
Can I ask for one too? I’d really like to see how the math stacks up to my own preconceptions.
If I’ve played enough games, that is.
Hey, looks like setting a maximum for player skill worked out pretty well!
In case anyone’s wondering why mirror matchup win rates have uncertainty again, as far as I can tell the model just ignores them, so it’s all good.
Ok, big dump of player/char skill charts. This covers all players who have played in the last year, or who have gotten to at least 1700 Elo.
Super cool. Can confirm, Fivec’s chart is hilarious.
while my chart might not have the perfect of Cpat Fivec Deluks or even Ntiiler, I’m proud of how mine looks. That is a LOT of characters that I can feel proud of.
Lol I just realized how bad my Lum was. Hoooly shit.
Charts look good. It really drilled home the point that most players are truly high level with just a few characters. My view of some other players skill was a bit more broad than the reality shows. As usual, your work is superior.
Honestly, I think your chart is just showing that you don’t play enough Lum for the model to have an opinion.
EDIT: I did some more digging, and I think the charts just need an adjustment. Right now, they try to make sure the area of each of the plots matches, so bars where they are more uncertain end up thinner. However, I think it’ll be more clear if they always have the same width (and then you can judge the confidence simply based on the distribution of values). I’ll make that change with the next round of them that I post. (I have a pending change/maybe improvement that tries to figure out what an average/unknown player has for player skill, and also weights games between experts more heavily than those between novices).
My chart shows me being semi-competent with half the cast. That’s what counterpicking does folks!