[Code S] Semi-Finals 2012 GSL Season 4 - Page 9

On October 09 2012 00:58 TheBB wrote:
I had some fun today.

Disappointingly, this forum doesn't have a monospace environment (that I know of), so my beautiful formatting is lost. (Or, I have to work to prettify it.)

Probabilities are based on the Elo ratings from TLPD, where equal weight was given to the total Elo and the matchup-specific Elo. It's not clear which weighing would yield the most accurate predictions.

Taeja, Life and Rain are top three in overall Elo. That makes Taeja vs. Life essentially the highest rated match possible at the moment. How often has it happened that the two Elo leaders met in a match while they were Elo leaders, and a best of five at that?

+ Show Spoiler +

Anyway:

Rain vs. Mvp

Rain wins
      3-0: 19.079%
      3-1: 24.287%
      3-2: 20.610%
      Total: 63.976%

Mvp wins
      3-0: 7.639%
      3-1: 13.194%
      3-2: 15.191%
      Total: 36.024%

Most likely winner: Rain (63.976%)
Most likely outcome: Rain 3-1 Mvp (24.287%)

Taeja vs Life

Taeja wins
      3-0: 22.514%
      3-1: 26.453%
      3-2: 20.721%
      Total: 69.688%

Life wins
      3-0: 6.008%
      3-1: 10.964%
      3-2: 13.340%
      Total: 30.312%

Most likely winner: Taeja (69.688%)
Most likely outcome: Taeja 3-1 Life (26.453%)

On October 09 2012 02:46 ColtraneL wrote:

You should precise the samples and on what they are based, because I suppose you're not using Rain's BO5 results in SC2 to do this because he only played two of them. Is it from broodwar?

On October 09 2012 03:13 TheBB wrote:

They're based on the Elo ratings. They, in turn, are based on performance in SC2 games only (not BW). Whether the games come from Bo5s or Bo3s or are standalone doesn't matter, they're weighted equally. In Rain's case, that's 60 games.

On October 09 2012 04:23 ColtraneL wrote:

Ok you did your maths this way. I don't think you can quite give the same value to games in a BO1 or BO3 series than games from a BO5/BO7 serie in those statistics. You basically give the same weight to games that are played in completely different situation. And yet again you need to mention the size of the sample, because except if you took only Mvp's 60 last games, you compare 316 games played over almost two years to the 60 games of Rain, that doesn't seem really meaningful to me.