When will littlegolem reach 2,000,000 games? General forum
33 replies. Last post: 20180418
Reply to this topic Return to forum
Carroll at 20171211
LG is over 1,900,000 games played, try to guess on which we will play the 2 millionth game ?
My guess is on 2nd of October 2018.

Carroll at 20171212
Should we agree on a limit date to post answers?
I propose 20180228 23:59:59 as this limit date.
Also maybe we should not accept manipulation by creating tons of new games on proposed date.

Carroll at 20171212
Or we could go until last second with a rating system giving more points for earlier answers depending on both distance to correct answer and distance in time for the prediction ?
I don’t know what the right formula for this point system would be. It seems like exponentially harder to give accurate prediction in advance, but maybe some estimators become precise after a few months of data and the months left until final time do not matter too much ?

Carroll at 20171212
Oh so another parameter to take into account for the rating formula is the precision of the guessing date ;)

purgency at 20171212
WHAT?!
Okay okay, i’m changing my guess. I have done all the maths. the last 100k games took 14 months. Given that there is only 90k to go it has to be approximately in 14*0.9=12.6 months. My guess is christmas 2018.

purgency at 20180301
TIMELIMIT WAS 20180228 23:59:59
UR LATE!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!

Carroll at 20180302
OK request granted, you have until 20180331 23:59:59
And I change my prediction to 20181205
Current predictions:
20180605 05:06:20.18: mmKall ? on 20171212
20180816: Ypercube on 20180301
20181205: Carroll on 20180302
20181225: Purgency on 20171212
There is still room for improvement...

Carroll at 20180303
It would be cool to continue to play near the time when it reaches 2M.
Anyone can come up with a rating formula giving more points to earlier predictions and accepting late predictions with less points when we approach the certainty?

wanderer_bot at 20180415
Where did you get that number? When I open the LG homepage I get a smaller number: 1,936,963 Games

ypercube at 20180415
I guess The_Burglar referred to the game with highest game_id and forgot that there have been some gaps in the id sequence.

William Fraser ★ at 20180415
I propose as the score sqrt(days between date estimate issued and date of estimate)  (days between date of estimate and date of actual).

Carroll at 20180416
I accept Bill’s proposition to have an inverse square law for predictions errors, with this law, here are the prediction budgets (in days) for each prediction received:
mmKall for 20180605 05:06:20.1: 13.23Ypercube for 20180816: 12.96
Carroll for 20181205: 16.67Purgency for 20181225: 19.44
 The_Burglar for 20180808 08:08:08: 15.49
The error in days will be subtracted from this prediction budget to get a score.

purgency at 20180416
So my prediction error has to be less than 19.44 in order for me to beat the dude that leaves his estimate on the very day that the gamecount hits 2 million, hence finishing with a score of 0?

William Fraser ★ at 20180417
Yep. I was assuming that at least one one competitor would get a score greater than zero. This system is probably not optimal for determining 2nd place, etc. Perhaps for second place finisher, the formula should double the bonus (meaning that you’d only need to be within 38.88 to beat that player).

Carroll at 20180417
Well, some theoretical work to backup formulas, constants would be nice.
I think the distribution of games played through time can be estimated.
The mean is easy, I have not tried to compute the standard deviation.
Knowing these figures we can compute how far an estimate may lie from the real date, 3 days before or ten days before or now and so change the bonus coefficient so that it is not way easier to get a positive score just the day before it reaches 2M...

Carroll at 20180417
As the distribution of games through time is a Poisson law (of varying parameter lambda, depending on new tournaments...), both the mean and the variance are lambda...

Carroll at 20180418
The probability distribution with Poisson l law that n games have been played on day t is given by the formula:
p = l.exp(l.t)(l.t)^(n1)/(n1)! which is a Gamma law.
With my estimate of l, that yields a 95% confidence that it is between d27 and d+27, with d best estimated date.
So the current budget is too low and I accept to multiply by 2:
mmKall for 20180605 05:06:20.1: 26.46
Ypercube for 20180816: 25.92
Carroll for 20181205: 33.33
Purgency for 20181225: 38.88
The_Burglar for 20180808 08:08:08: 30.98