Page 1 of 1

Rounding 3bvs scores

Posted: Mon Dec 06, 2010 7:17 pm
by sleep
Clone gives 3bvs scores with three decimal places and the other versions give two decimal places. Every version rounds the scores normally, for example if your accurate 3bvs is 3.5467, the counters window will give 3.547 on Clone and 3.55 on MSX, Arbiter and Vienna.

As you might guess, in 3bvs World Ranking there are scores among Arbiter and MSX games that are have been rounded up, for example Manu's Beg 7.166->7.17 and [name removed]'s Exp 4.256->4.26. However, all the Clone scores are rounded down, regardless of what is in the third decimal place, for example Kamil's 4.897 is 4.89. Am I wrong or does this issue need fixing?

Re: Rounding 3bvs scores

Posted: Mon Dec 06, 2010 10:17 pm
by EWQMinesweeper
unofficial response:

clone-users fault

rest: self-censorship :(

official response: had that on IRC not long ago, will be fixed soner or later

personal note: there are more important things to talk about than rounding hundreths!

[closed]

Re: Rounding 3bvs scores

Posted: Sat Feb 05, 2011 9:43 pm
by thefinerminer
As some people say, 0.01 is not very important. I investigated this in December 2010 after 'sleep' (Tomi) emailed me. Results:

Arbiter:
0.47+ cuts (10.666-->10.66). Earlier versions rounded.
The reason the ranking sometimes has different decimals than the video is because early versions rounded but the ranking cuts.

VSweep:
The history rounds down (2 decimals), the counter rounds up (2 decimals), and the filename gives 3 decimals but I do not know how it rounds. Christoph has said many times that the time on all clones is accurate only to +/- 0.004 so 3 decimals does not mean anything. For the ranking I cut to 2 decimals.

Clone:
All clone versions seem to round. I cut Clone2007 to 2 decimals.

MSX:
All versions seem to round.

Result:
Arbiter/VSweep/Clone2006/Clone2007 stats can be compared exactly because they do not round or I cut them to 2 decimals.
Clone97/MSX round - this means half the time the 3bv/s is 0.01 better than it should be, and half the time the score is 0.01 slower than it should be.