@marcelpclash β 0.065% chance to have only hit 1 Sivir by that point
(Some sampling bias because you chose right before you hit one, but even then it was a 0.2% low roll on 2* Sivir)
@MikaMikaX_@MetaTft Not ToS, but having people force the no. 1 comp kind of defeats the purpose of PBE.
By presenting info like this people can get an idea of what is good, whilst also encouraging comp diversity as we put low playrate comps towards the top
@Mortdog Hi Mort, glad you're recovered!
Can I ask what the current pool sizes are?
Datamining seems to suggest 22/20/17/10/9 but I haven't seen a change from Set 12's pool sizes announced anywhere
Somewhere to see this info in the UI would be great too
@CuewarsTaner@MetaTft Yeah you can only paste team codes when out of game, but otherwise it looks pretty similar in game
So you can load up a bunch of your favourite comps before game, then toggle which one is active while in game.
Its live on PBE if you want to test it out π
@BrosephTft The comp does say to play Xerath > Ryze
Ryze is played on Lvl 8 boards but is then swapped out for Xerath or Milio as you say.
That creates survivorship bias as Ryze only appears on boards that don't hit, so he probably performs better than the stats might indicate
@AesahTFT Does this mean a crest is bad if you're playing Kalista? Not necessarily, the 3-2 stats are good.
I think its very difficult to compare the performance of a linear option like a crest, with a flex option like Combat Caster or Scoreboard Scrapper (specifically at 2-1)
@AesahTFT Another interesting factor here is that EVERY crest has a worse placement at 2-1 vs 3-2.
My intuition says this is because a crest at 2-1 forces you into a direction, whereas a crest at 3-2 is tailored to your board and aligned to your direction.
@Abner_sena_@MetaTft Sure! It's a Tensorflow model trained on 2,000,000+ individual rounds from app users across all stages of the game.
It takes into account the champs, their star level and items, active traits and augments (but not positioning), and is 84% accurate when tested on real games