ISML Fantasy: Sapphire

[imbehindyou]

I just noticed that Eater-of-All, me, and kiero all have rating of 90.0855

[Team Rocket Elite]

I created initial standings for all returning characters based off their ISML 2008 performance. I created Extrapolated Standings using direct performances against several high tier character and converted the scores into expected percentage against Fate, then took an average. For new characters, I used their preliminary round performance against characters from ISML 2008. The characters this doesn't work since they had 0 matches before the start of Aquamarine were Kirino, Akari and Yami. I just guessed for those three. I went on to win the Aquamarine fantasy despite a incorrect 1 character necklace prediction so this apparently worked pretty well.

I use these scores to generate predictions for each match. After each match, I lower the scores of characters that perform worse than expected and increase the scores of the characters that did better than expected. Larger under/overperformances lead to larger swings in rating.

The formula I use to create predictions is:
[(C's score)/(B's Score)]*50 = (C's percentage against B)
Where C is the weaker character and B is the stronger character.

It's simple but summarises how people vote pretty accurately. I guess it also kinda implys that voters are mindless drones that can be summed up with 2 variables, 2 operations and a constant which I guess some people might not be happy about. >_>

To create my Fantasy prediction, I use my latest ranking from the end of the previous match. This ignores changes due to new releases but i'm too lazy to bother with those effects. <_<

The Extrapolated Standings method assumes that the character strength is linear and transitive. SFF matches throw out the linear part, but it doesn't change the winner so it doesn't matter too much for Fantasy predictions. SFF matches just needs to be thrown out when adjusting between matches. Rallying and anti-votes aren't accounted for but they are pretty negliable factors. Cheated votes completely throw the system out of whack if they aren't caught. The system doesn't work as well for polls with low vote totals since normally small factors are no longer negliable. Some voters leaving or new voters coming in doesn't change much since the people leaving / coming in don't really vote any differently from the people currently voting.

[kiero]

Wow, complicated.
My method is a lot more programmatic (with a lot of human adjustments);
I just try to connect each character to her opponent in a big graph, and calculate the relative strengths of the characters.

For example, if I was trying to predict Tomoyo vs Suigintou, and I see that Suigintou had votes of a factor of 1.3 over Hina and Tomoyo had votes of a factor 1.2 over Hina, Tomoyo's strength vs Suigintou would be 1.2/1.3, and in this case, it would predict she would lose. Of course, there are a lot of possible comparisons, but the comparisons with "shorter paths" take greater weight in the predictions.

It does this based on last period's statistics for every character, taking about 30 minutes in total.
Once that's done for all the matches, I look over them and change anything that I think might be wrong, and this hasn't done too badly the last two periods (I did it differently in Aquamarine).

[walyi]

I make a lot of mistakes at this race
OTL

[Fate]

Thanks for the tips guys. It is rather complex...I guess I'll just stick to my intuition. o_O

[Juice]

I use my own rating system (which USED to be published). It bases it off points, SDO, and percentage of VF and VD. I then use these values along the KS script for fantasy to create a script that outputs 1s and 0s. This new script then outputs the sum of these 1s and 0s for each Contestant in ISML. Needless to say, I despair quite often at the results of this output. T__T

[minhtam1638]

Mine is based almost solely on Average VF. If a candidate has 100 more Average VF than her opponent, that candidate automatically wins in my opinion, so I don't waste time with those matches that have that kind of a difference. If the difference between average VF is less than 100, that's where I go with my gut and look at the following trends.

[Demi.]

100VF is such a tiny number with how many votes ISML is pulling in. And what about SDO?

Edit: Oh, do you mean average VF as in, 100 more VF per match?

[Chocola]

Man minty you have no life. Spend a bit helping us actually run the tournament.

[Cecaniah]

Oh, do you mean average VF as in, 100 more VF per match?

This. Why do you think I was fourth and thrid in the last two?

I may not be the best, but damn good.

I use the votes to get the predictions.

However, since last period was fucked I couldn't use that.

[Fate]

I wonder how much points I lost. --;

[Chocola]

After Sapphire 4:



It seems that in every round, someone pulls a Saki and get +/-0.

[Eater-of-All]

Wowowow, I lost exactly what I gained last round.
I hate life.

[Demi.]

Nice rise, too bad I was being retarded when I made my predictions.

[imbehindyou]

Hey, I went up!

3 wrong predictions:
Holo vs Ibuki Fuuko
Anya Alstreim vs Ryougi Shiki
Ryuugu Rena vs Kawasumi Mai

[Alexander]

I keep losing points

[walyi]

The 15th place......

Oh it doesn't matter

[Cecaniah]

I'm only 13th because everyone above me is getting better.

I rose.

[Sakurei]

wow that was a heavy blow xD

[Fate]

I lost quite a few too...