A New Bug?

Case number:699969-2003941
Topic:General
Opened by:Merf
Status:Open
Type:Suggestion
Opened on:Saturday, July 15, 2017 - 17:51
Last modified:Tuesday, August 29, 2017 - 15:22
(Sat, 07/15/2017 - 17:51  |  7 comments)


Merf's picture
User offline. Last seen 16 hours 23 min ago. Offline
Joined: 01/01/2010
Groups: None

Hello, I think I've found a brand new Bug! I haven't patented it so if anyone
wants to try it, feel free.

Puzzle 1404

When I select "Restore Very Best" I get my protein completely cut up, as though
I had selected the macro "Cut and Wiggle Everything". I have no idea why this
has happened, I've tried logging out, and that hasn't cleared it. Any ideas anyone?

Cordially,

Merf.

+ Apologies for previously posting this elsewhere!

LociOiling's picture
User offline. Last seen 1 hour 26 min ago. Offline
Joined: 12/27/2012
Groups: Beta Folders

I'm not seeing this behavior on Puzzle 1404 on Windows.

"Restore Very Best" restores your best solution, even if there are problems like open cutpoints. Sometimes a solution will be better with a open cutpoints.

Also see recentbest ignores residue count penalty. In recent design puzzles, the very best ("abolutebest") and credit best ("creditbest") poses may incorrectly fail to deduct the penalty for added segments.

For example, I have a very best on 1404 that shows as 10,751 in open/share solutions with "show auto and quick saves" checked. This solution loads as 10,291. The difference of 460 points is the penalty for 20 added segments, at -23 point per segment.

So there are a couple of reasons the automatically saved poses may not be accurate.

LociOiling's picture
User offline. Last seen 1 hour 26 min ago. Offline
Joined: 12/27/2012
Groups: Beta Folders

Oops, named the wrong pose. The recent best ("recentbest") pose is affected by the segment penalty bug, along with absolute best. The credit best pose is not affected by this bug, but there have been cases where a credit best pose has a lower score when you restore it.

Merf's picture
User offline. Last seen 16 hours 23 min ago. Offline
Joined: 01/01/2010
Groups: None

Hello again,

Thank you Loci for your input, I have now found the workaround for this
problem following your credit best note.
The mystery is that when I select "very best", the score does
not indicate MY very best, but something well beyond, that I have not
attained. Perhaps there is some bleedthrough here from somewhere else but
I have certainly been puzzled to explain this. Is the "system" making kindly
suggestions as to which way I should go?

Cordially to all,

Merf.

LociOiling's picture
User offline. Last seen 1 hour 26 min ago. Offline
Joined: 12/27/2012
Groups: Beta Folders

Sadly, it's not the system making helpful suggestions.

The problem is a scoring bug which fails to deduct the penalty for added segments. This bug affects the recent best and very best poses on design puzzles.

To see this for yourself, in Open/Share Solutions (control+O), check "Show auto and quick saves" on 1404 or another recent design puzzle. Under "Local solutions", you should see an "Autosaved Very Best Soloist Position". If you've added segments this "very best" score should be higher than your current score.

For example, on 1404, I have a solution that's scoring 10,548. The very best solution shows 11,008, a difference of 460 points.

When I open the filter details by clicking on the triangular "conditions satisfied" dropdown and scroll down, I see:

Residue count: lost points (-460)
105 residues found (20 over limit of 85)

highlighted in red. This penalty is not being included in the score for the "very best" or "recent best" poses.

It's possible that the same problem may occur with other filters, in particular the ideal loops filter. I'll look into that on the next design puzzle.

Here's what this example looks like:

residue count penalty bug: The penalty for added segments is not included in very best and recent best poses.residue count penalty bug: The penalty for added segments is not included in very best and recent best poses.

Merf's picture
User offline. Last seen 16 hours 23 min ago. Offline
Joined: 01/01/2010
Groups: None

Gosh! Thank you Loci for this comprehensive piece of work.
I think that I now know where the original problem lay, and
how it meshes withe your explanation. I have not experienced
this kind off thing before, and was very curious as to its
source. Now I know. What a friendly place Foldit is!

Cordially,

Merf.

Joined: 09/24/2012
Groups: Go Science

Hello,

Thanks Loci for the explanation. It appears to be a bug.

It might explain why some of my recipes can "loose" points on filtered ideal loops. This happens with the option "generic filter" written by BitsPawn and that you and me included in a couple of recipes. In principle, most actions occur with filters disabled. When a score is calculated, the filters are first enabled. But with this bug, when calculating "recentBest", the filter stays disabled. Then the recipe thinks there is an improvement and goes on with a lower scoring solution.

I thought I fixed this problem revealed by gitwut, but I saw it again when an unideal loop penalty appeared in a design puzzle. I really didn't understand the source of my recipes "bug".

Now I understand that this is a Foldit bug !!

See here:
http://fold.it/portal/recipe/100682#comment-35054

Thanks again Loci and Merf for discussing it.

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