statsmash

Summit 2 Projections


Tweet

literal mountain summit

Summit 2 comes up tomorrow, and with it a mix of the best and the pretty good. To be honest, we already know what the model thinks of the players, but it’s an interesting exercise nonetheless.

First, let’s talk about the groups. Since they’re all round robin, no need really to do the calculations;
it’ll just be in descending order of our rankings. So here are the project group results

Group 1 Group 2 Group 3 Group 4
MkLeo Samsora Tweek Nairo
Light Marss Dabuz Tea
ESAM Maister zackray VoiD
Armada RFang MuteAce Leffen

Now on to the juicier stuff: end results. We’re going to run the bracket 500,000 times, picking winners and losers for each match probabilistically based on our Elo scores, and then see how many times out of those 500,000 trials each player got what result. Monte-carlo sampling.

1 2 3 4 6 8 12 16
MkLeo 0.54552 0.69152 0.77636 0.83762 0.91058 0.95638 0.99072 1.0
Tweek 0.12422 0.33772 0.50858 0.64114 0.80302 0.89152 0.98542 1.0
Samsora 0.10284 0.28072 0.42774 0.54980 0.71024 0.80972 0.95078 1.0
Nairo 0.08974 0.21772 0.35112 0.47936 0.68894 0.79544 0.94748 1.0
Marss 0.04460 0.11494 0.19366 0.28230 0.48762 0.70492 0.94302 1.0
Dabuz 0.02726 0.08104 0.15638 0.25070 0.45588 0.61182 0.89780 1.0
zackray 0.01948 0.07856 0.15434 0.23310 0.39740 0.56870 0.87440 1.0
Light 0.01906 0.07680 0.15106 0.22978 0.39518 0.56944 0.87730 1.0
Maister 0.01584 0.06698 0.14118 0.21834 0.41038 0.61072 0.95628 1.0
ESAM 0.00806 0.03106 0.06996 0.13112 0.28820 0.44858 0.80978 1.0
Tea 0.00182 0.01344 0.03814 0.07160 0.18976 0.38284 0.88388 1.0
VoiD 0.00120 0.00628 0.01826 0.04202 0.12776 0.31422 0.70490 1.0
Leffen 0.00014 0.00110 0.00490 0.01036 0.03846 0.08862 0.29752 1.0
MuteAce 0.00014 0.00108 0.00352 0.00990 0.03798 0.11592 0.36136 1.0
RFang 0.00008 0.00096 0.00432 0.01190 0.05124 0.10768 0.34494 1.0
Armada 0.00000 0.00008 0.00048 0.00096 0.00736 0.02348 0.17442 1.0

First thing I notice at least is that the model treats Leo like an absolute God, with a MAJORITY win-probability, which is absolutely insane in a 16 man bracket.

But what’s perhaps the most shocking is Leffen: for some reason, despite straight up being the lowest ranked besides Armada, he has a higher chance of winning than Muteace and RFang.

Poor Armada did not win a single time out of 500,000 simulations. In this case, his lack of attendance probably means we’re greatly underestimating his strength, but sadly even on a eye test he does seem to be the worse of the attendees.

There is interestingly a gulf between Nairo and the rest of the competitiors; after him, no one is even close to 10% win probability.


On an unrelated note, the other project I’ve been working will be out (in alpha form) sometime the next few weeks if all goes well. That project is basically “Pickems”, which should be familiar to Dota, League, or CS:GO players, with a future expansion into some kind of fantasy sport addition as well. Naturally, no gambling here, everything will be free to enter with probably no reward or minor reward.

Contact me at stu2b@statsmash.io or @stu2b50 on Twitter

← Back