Wrong use of SPRT

Skip to first unread message

Uri Blass

Feb 9, 2018, 2:38:39 AM2/9/18
to FishCooking
I read in the comments for tests that somebody want to check a scaling of the contempt. C=25 failed LTC non-regression. Can it pass STC?

If people want to test scaling then SPRT is useless and they need to use fixed number of games because SPRT unlike fixed number of games tell almost no information about the difference in playing strength.

using 100,000 games at STC and 100,000 games at LTC will give us better information about scaling if this is the target.
I doubt if we will find a significant difference and results of SPRT can only cause some illusions for example in case that the probability to pass SPRT is 50% both in STC and LTC and in one case it fail SPRT and in one case it pass SPRT.



Feb 9, 2018, 4:19:42 PM2/9/18
to FishCooking
пʼятниця, 9 лютого 2018 р. 09:38:39 UTC+2 користувач Uri Blass написав:
Agree with you.

Alexandre Meirelles

Feb 9, 2018, 5:46:58 PM2/9/18
to FishCooking
I agree too.
Reply all
Reply to author
0 new messages