RoRoRo and predictions

Just in case you've not noticed, RoRoRo is attempting to automatically create a forum thread for each new championship's top league. This includes variants where there's sufficiently high turnout (several leagues).

So hopefully nobody will need to do the requests for these any more. Unfortunately, one did slip through the net and not have predictions posted, then it's probably best to message me rather than attempt to start them yourself, as if you start them it will change RoRoRo's internal state, and it may be harder for me to debug.

As a side note, here is the ratio of games which he has found worthy of comment on:

8046 game results have been seen
190 (2.36%) were considered upsets
83 (1.03%) were considered major upsets
15 (0.19%) were considered astonishing.
So basically 1 game in ~28 got some kind of comment.

I'm fairly happy with those ratios. If you've had what you thought was a heroic victory and he didn't make an appropriate comment on it, or if you think he's getting too overexcited too easily, please let me know.

I would like to see him upset or something similiar if the game was "the" decision in a championship.

That's a hard one. If Johny and Kfieco, say, are fighting for 1st and 2nd place but have finished their own game, then it might be a game between me and RoRoRo which flips the result from undecided to decided, no matter how tedious our result might be. Two results may come in at the same time, which one is responsible?

RoRoRo tries to minimise the amount of state that he maintains, but one of the flags that he does store is whether the championship's decided (he won't post after that, you may have noticed). Therefore he can do something different when that flag changes state.

If you can come up with something more concrete and algorithmic, I'm all ears.

Well, isn't it simply exactly the game where p(won) in first place matches 100% for the first time?

But if u do not want ton run him more often when more than one game is decided it would be nice to annouce the decision in any way.

Just for the record... the game that triggers the 100% for the first time could be undecisive for a another reason... there is a chance smaller than 1/10000 that the playere with 100% is not the later winner :-)

A game that chances the p(won) from 99,7 to 100 is hardly crucial!

Call a game "crucial" if it changes the outcome of p(won) for at least 1 player with more than 10%.

I really like Hjallti's idea of a big swing detector, but alas that requires memorising previous predictions, which he currently doesn't do. (minimising state, as always.) I'll think about it, but I have about 20 other things in the queue currently, we'll see how it fits in.

Can't Rororo not "take" the previous prediction from the forum?

And getting back to the original post: word games, gomoku and connect6 are not yet included in the lot of automatically created prediction forums. Is it only because no new championships have started recently? Both wyps.ch.13.1.1 and gwg.ch.25.1.1 seem to be wide open for predictions (although connect6.ch.14.1.1 has not yet started because of Richard's vacation).

Should I trigger predictions for these?

And how do you trigger a forum thread with RoRoRo's predicitions? I am curious so I give it a try.

> alas that requires memorising previous predictions

How about

(A) at the moment the champion, X, is determined (I don't know if RoRoRo has code to recognize this, but it seems to me it would be less of an effort than running 10,000 simulations)

(B) go through all the then-completed games one by one and

(C) calculate the projected probability of X becoming champion as you would if the selected game had not yet finished.

(D) Declare the game most crucial to X's victory as the one where the probability calculated in C is lowest.

No memory of past predictions or of the order in which games finished is necessary.

Step C looks like it could require 10,000 simulations for each of the completed games. That sounds like a lot, but it's no more than RoRoRo may have done already in making his ordinary predictions. And since the number of outcomes ((2 or 3 for most games, depending on whether draws are possible)**(number of not-yet-completed games plus 1)) may well be less than 10,000, that number may be substantially reduced.

MG - his sims are 50000 games (all done in perl, it can take over a minute sometimes). I wouldn't want to do 40-200 (street soccer) of those, even if I cranked it down to 10000.

bloke - send him a message with the subject line "predict help"

@Phil, Upon further thought, steps B and C can be done in at most GAMES*SIMULATIONS operations, where GAMES is the number of games in the tournament and SIMULATIONS is the 50,000 number [rather than requiring (completed games)*(uncompleted games)*SIMULATIONS ops]. That's the same number of ops as currently used to generate one set of estimated probabilities early in the tournament of the various players becoming champion.

Step A can surely recognize whether any player has clinched the tournament in GAMES*PLAYERS*PLAYERS ops, and I think in at most GAMES+PLAYERS ops [e.g. in GAMES ops calculate players' current points from completed games and the points possible based on uncompleted games, then in PLAYERS ops check whether any player is unassailable], either of which is trivial compared to GAMES*SIMULATIONS.

This is not to say that identifying a most-crucial game is worth it.