691 scoring anomaly

Case number:671071-994897
Topic:Game: Other
Opened by:Bletchley Park
Opened on:Thursday, April 4, 2013 - 23:37
Last modified:Friday, April 5, 2013 - 17:48

I have a client running with a max score of 17311. I moved into its track with another client and checked the saved quicksaves and autosaves. They show 17311 as a score. When I load these quicksaves and autosaves, I get a 17111 score. This seems a bug to me.

I have saved one such reloaded solution as oddscore311is111 shared with myself in Contenders group save.

I'm running 2013-03-26-54492:54524M-win_x86-devprev

(Thu, 04/04/2013 - 23:37  |  7 comments)

Joined: 05/19/2009
Groups: Contenders

And to clarify, these are all filtered scores.

Joined: 09/21/2011
Groups: Void Crushers

Bletchey, do you see a penalty and missed bonus in the 17111 score?
That might add up to 200 and then it is the issue that the restored position is just enough different from the original ingame position that the penalty is issued. (I think that the precission of restoring a position from a savefile is almost never exactly 100%, and the filters are very sensitive)

Hanto's picture
User offline. Last seen 7 hours 31 min ago. Offline
Joined: 05/10/2008
Groups: None

It would seem to me that the save file should be more exact then any filter's sensitivity, if not it lends itself to far to much erroneous behavior. How could the save be considered a digital save and be so inaccurate? Voting UP

Joined: 09/21/2011
Groups: Void Crushers

Hanto, floating point arithmetic is never exact, there is allways a cut-off point in precission. And the filters can be very sensitive to even very tiny changes in position, so a segment considered to be core in the save might be not in the core in the restored version. This problem is in my opinion inherent to the use of floating point arithmetic and filters that use fixed requirements.

bertro's picture
User offline. Last seen 1 week 2 days ago. Offline
Joined: 05/02/2011
Groups: Beta Folders

then this error should go both ways, down as well as up sometimes?

Joined: 05/19/2009
Groups: Contenders

At first I dismissed the idea of the tiny difference accounting for this issue. I then gave Timo credit and tested it. Low and behold, after twitching it and letting it wiggle, the condition was met, the -40 plus the 160 bonus added up and the score was 311 again..

Question then remains, should this be seen as a bug ? Should this be rounded differently or saved differently to correctly reflect what was saved ?

(and thank you Timo for taking the time and effor to explain it).

Joined: 09/21/2011
Groups: Void Crushers

Bletchey no problem. I have had the same issue on another puzzle. See: http://fold.it/portal/node/994713


Developed by: UW Center for Game Science, UW Institute for Protein Design, Northeastern University, Vanderbilt University Meiler Lab, UC Davis
Supported by: DARPA, NSF, NIH, HHMI, Amazon, Microsoft, Adobe, RosettaCommons