Strange wiggle behavior

Case number:699969-988639
Topic:General
Opened by:GaryForbis
Status:Closed
Type:Question
Opened on:Friday, October 8, 2010 - 15:36
Last modified:Saturday, July 27, 2013 - 04:32

OK, I've only been playing FoldIt for a couple of months now so I don't know the range of normal behavior in various puzzles. I don't know if what I'm about to describe is a bug or a feature.

In the past I noticed that do_global_wiggle_all(1) ran for different lengths of time but this seemed to be related to the territory being wiggled. I assumed that this was because FoldIt used some indexing scheme where it didn't have to do all the calculations but could just follow positions already calculated if they existed.

For about the last week, week and a half, wiggle seems to have one behavior when 50 or more points from my current best and another once it reaches that range. The score moves regularly both up and down when well below current best but will stop abruptly for a bit then jump then stop then jump, etc. when close to my current best. I need to do some tests but it appears like

for x=1,5 do
do_global_wiggle_all(1)
end

might produce different results than

do_global_wiggle_all(5)

On a few instances I done a wiggle all through the user interface and moved the clashing from 1.0 to 0.0
and seen no movement in score or protein position. Does this kind of behavior seem reasonable on 369 and killer toxin when well off the best scores? I don't think I'm in a local minima. As I said it seems to depend upon my current best score. When my best moves up so does the behaivor. Maybe I really am moving from minima to minima without knowing it. My intuition tells me that if minima are easy to find then they should be independent of relative score.

I live in a somewhat paranoid world so I have to discount my own feelings on the matter.

I've taken to using a modified wiggle routine with lopsided behavior:

x=do_global_wiggle_all
function do_global_wiggle_all (i)
local delta=
local score
local newscore=get_score(true)
repeat
score=newscore
for j=1,i do
x(1)
end
newscore =get_score(true)
until newscore < score+delta
end

That seems like a lot of work just to fight paranoia.
It still depends upon a honest wiggle.
I'll modify more to find the better mix.
Heck, I may mix backbone and sidechains in.

(Fri, 10/08/2010 - 15:36  |  3 comments)


Joined: 09/18/2009
Groups: SETI.Germany

Yes, I noticed that, too.

A wiggle of 1 runs longer if the backbones are not that settled (after alignment or some reset positions), or if you use many or strong bands.

If the puzzle is settled or there are no (strong) bands, it definitely runs shorter.

Joined: 09/18/2009
Groups: SETI.Germany

My thoughts are similar.
I think, the possible range is calculated once each time before wiggle is executed, and the number iterations selects a fraction of this range of movement.

So, by calling wiggle 5 times, the range of movement is also checked 5 times.
Performing a single wiggle of 5 iterations, the range of movement is only checked once.

spmm's picture
User offline. Last seen 28 min 43 sec ago. Offline
Joined: 08/05/2010
Groups: Void Crushers
Status: Open » Closed

time bound

Sitemap

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