puzzle 1590 wild scoring

Case number:699969-2006116
Opened by:LociOiling
Opened on:Saturday, October 20, 2018 - 01:22
Last modified:Monday, October 22, 2018 - 05:19

Several players have scored over a million points on puzzle 1590.

The puzzle has been made worth zero points and moved to the devprev category.

Using 20181018-f195fe62d9-win_x86-devprev, I took the starting position and ran Mutate No Wiggle on low wiggle power, followed by rav3n_pl Fuzes, also on low.

Switching to high wiggle power, I ran a modified Acid Tweeker on high wiggle power. There was soon a huge jump:

Gained 2.501 on seg 117 (17/40) scoring: 16.173, total score = 11228.749
No gain up to seg 112 (18/40), score = 11228.749
Gained 1376059.523 on seg 8 (19/40) scoring: -6578.193, total score = 1387288.273
Gained 2.719 on seg 192 (20/40) scoring: 5.106, total score = 1387290.992

There was nothing interesting in log.txt, lots of "no proper DoF" messages, and a few "response code 413 from server" while we where having server problems. The following lines occurred twice:

WARNING: residue 125 atom 4 input stub atoms nearly colinear at 1.73043 degrees
WARNING: residue 172 atom 4 input stub atoms nearly colinear at 1.59848 degrees
WARNING: residue 182 atom 4 input stub atoms nearly colinear at 3.23884 degrees

I'll try again, skipping directly to high wiggle power. I've saved log.txt and the scriptlog, so they're available if needed.

(Sat, 10/20/2018 - 01:22  |  9 comments)

Hanto's picture
User offline. Last seen 30 weeks 6 days ago. Offline
Joined: 05/10/2008
Groups: None

Check the puzzle comments please, You will see the FIRST troubleshooting there.

Hanto's picture
User offline. Last seen 30 weeks 6 days ago. Offline
Joined: 05/10/2008
Groups: None

IF you are going to be changing the status off a puzzle it should show up in the main client TOP CENTER. BY doing what youhave done, valuable time, resources, AND EFFORTS have been wasted. NO scoring BUG, but as usual DEVS do not QA the puzzles AT ALL.

Hanto's picture
User offline. Last seen 30 weeks 6 days ago. Offline
Joined: 05/10/2008
Groups: None

AT my current #2 position on the puzzle, a LOT of crashing is happening but a lot of different factors could cause that. THE worst problem as stated in puzzle comments is segment 13 which responds with 1.1x10 to the 6th power when hit with a default glycine hinge from puzzle opening position in AUTO. APPARENTLY locked segments WERE NOT LOCKED as they should have been.

Joined: 12/27/2012

Based on the previous comments, here is a quick procedure to recreate the scoring problem:

1. Open the puzzle, reset, create a new track
2. Set medium wiggle power
3. Download the KarenCH version of glycine hinge: https://fold.it/portal/recipe/100513
4. Run the recipe, clearing the hinges box and entering "6".

The score quickly jumps to over a million points.

Logs and scriptlogs available if needed, but these steps seem to work reliably.

bkoep's picture
User offline. Last seen 2 days 10 hours ago. Offline
Joined: 11/15/2012
Groups: Foldit Staff

Thanks, Loci! This is really helpful feedback; simple steps to reproduce the problem will make this bug much easier to track down.

We'll still have to dig through the recipe to see exactly which step is responsible, but this is a great starting point! (If anyone can reproduce the problem without a recipe, or with a minimal single-function recipe, that would be even better!)

Joined: 12/27/2012

nbl_hinge 1.5 will also produce the same effect: https://fold.it/portal/recipe/100081

Use the steps above, start nbl_hinge with default options, let it run.

My results:

NBL_hinge v.1.5
Puzzle: [DEVPREV] Aflatoxin Challenge: Round 11
Track: default
number of runs = 1
hinge width = 1
random order = false
loops only = false
segment range(s) = 3-198
Start, score: -21014.686, Sat Oct 20 23:40:36 2018
hinge run 1/1, score: -21014.686, total gain: 0, Sat Oct 20 23:40:36 2018
CI = 0.3, band strength = 0.25
segment: 3, gain = 1116786.953, score = 1095772.267
segment: 4, gain = 86729.837, score = 1182502.104
segment: 5, gain = 132290.624, score = 1314792.728
segment: 6, gain = 11938.919, score = 1326731.648
segment: 7, gain = 9945.301, score = 1336676.95
segment: 8, gain = 11489.1, score = 1348166.05
segment: 9, gain = 4557.149, score = 1352723.2
segment: 10, gain = 7288.53, score = 1360011.731
segment: 11, gain = 2589.51, score = 1362601.241
segment: 12, gain = 2135.273, score = 1364736.515
segment: 13, gain = 183.413, score = 1364919.928
no gain after segment: 17
no gain after segment: 21
no gain after segment: 26
no gain after segment: 31
NBL_hinge v.1.5 cancelled
Puzzle: [DEVPREV] Aflatoxin Challenge: Round 11
Track: default
Final, score: 1364919.928, gain: 1385934.614, Sun Oct 21 00:33:45 2018
Final, hours: 0.885, points per hour: 1564554.597
Joined: 12/27/2012

Not sure if this image will be readable, but segments 13 and 14 saw ideality gains of about 750,000 points each in the previous test, which used nbl_hinge on the starting pose.

Segments 2 through 12 showed significant ideality losses in the same test.

The test was cancelled somewhere after segment 31, so the whole protein was not affected.

Joined: 12/27/2012

One more try -- condensed version of the above.

Hanto's picture
User offline. Last seen 30 weeks 6 days ago. Offline
Joined: 05/10/2008
Groups: None

Thanks Loci, Id upvote if i could see to do so. This blue screening of typed comments is actually quite detrimental to what little vision I have.
Now an update, crashing was only evident after reaching the 1522000 area of operation on my original client on an A10 processor. Once changed to an intel crashing has stopped ( so far for 12 hours, knock on wood ). Progress has been slow with me still operating AUTO mode and no attempts YET to address segments 200-202. theres at least 35-40,000 points left in the puzzle, probably much more then that for a good folder. I''m leaving for Manilla in a few hours to be gone for near 3 days. This computer with me of course, but travels in Manilla are exhausting for me, so my attention will not be here. Still have not any example of wild scoring, different maybe, but not wild. Wish i had more to say, but you now have all that i can validate. Cheers


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, Boehringer Ingelheim, RosettaCommons