better prediction for running tournaments?

[zxexuxs]

For example now during the dreamhack bo16 in the single elimintion bracket: TLO (1431) won against squirtle (1920) with 2-1 This can mean 2 diffrent cases. First glicko rating is right but the improbable case have occured. Second, glicko rating was not accurate. Maybe TLO is in a really good shape so his rating for this tournament have to be higher, or squirtle's lower.

Couldn't the prediction be enhanced by immediatley updating the rating? So TLO games get calculated with a higher rating in the further tournament.

If i get it right, the consistensy of the player influence the winrate. Then a modification of this value instead of the rating could enhance the prediction, too.

I hope you understand my idea.

[TheBB]

So the suggested fix is probably something along these lines:

• When a match is changed, added or deleted in the most recent period, update the ratings of the two players involved.
• On changes in past periods, do nothing (since these will be quite expensive).

It's a possibility, but it's not something I consider a huge priority, since the rating adjustments from a single match generally have negligible impact on the predictions.

[zxexuxs]

"a single match generally have negligible impact" Yes thats right but if there would be a prediction which is adaptable to various tournaments, this is getting more interesting. With adaptable i mean that it is possible to make a prediction with 2 or 3 group stages which followed by an single elimination bracket or double elimination, etc. Then you can follow an underdog about 4 or 5 matches...

[TheBB]

On account of the reasons given above, I consider this not worth it. Closing.