"Wiggle all" not variable enough at low Clash

Case number:845813-991916
Topic:Game: Tools
Opened by:bareego
Opened on:Sunday, February 19, 2012 - 06:40
Last modified:Wednesday, February 22, 2012 - 19:35

The wiggle all function or the Clash slider do not the work the way
they used to and the way they IMO should.

Do this experiment.
Take a protein, turn on wiggle all at Cl 1.
It will settle for a value quickly.
Now move the slider down, it will start to go down but even at low Clash values (eg 0.2)
the score will quickly settle at a point and not or hardly move at all.
This is not good.
You want things to change a lot at low Clash.

(Sun, 02/19/2012 - 06:40  |  11 comments)

Joined: 06/17/2010

Looks like Ci=1 wiggle is "too strong".
Then lowering CI not make much difference somehow.
Also: there is HUGE difference when we use "all loop" mode vs "structure" mode on scripts. Wiggle is reacting "too much" now. "Structured" protein is way too stiff.

Joined: 06/17/2010

Script: http://fold.it/portal/recipe/38768

Puzzle: Beginner Puzzle: Anthrax Lethal Factor Protein
State: from reset.

Result: NO CHANGE on wiggle_all(1) and (2). (3) and more making changes.
BUT it is MUCH LESS change that we have in old clients!

Video: http://youtu.be/scOSSUABVBM

Joined: 04/19/2009

The current client has somehow killed endgame - it doesn't really exist right now, walking will give fractions or a couple points only. Somehow, something is stiffening or changing the protein at late mid-game.

This is what led to trying to test wiggle & CI. I did some testing with the rav script above on a reset of the Anthrax puzzle, and used an old client to compare with the current. I had to modify the do_global_wiggle_all to (3) for the script to show any changes on my mac (possibly sticky mac clients?).

The old client from before the 10-6 update ran twice as fast as the current client, and the results were very different. It appears that the current client is not quite a rollback to this older client, which was the one before the introduction of the WA slider.

Old client:

Version 207f7fae2b5b034e1a5fcad06383f6a6

.01 = 147.905
.02 = 5355.625
.03 = 6390.207
.04 = 7214.516
.05 = 7581.453
.06 = 7925.008
.07 = 8062.536
.08 = 8870.871

Current client:

.01 = 797.750
.02 = 4927.987
.03 = 6315.994
.04 = 7344.077
.05 = 7631.437
.06 = 7871.987
.08 = 8243.676
.09 = 8307.467
.1 = 8870.830

spmm's picture
User offline. Last seen 19 weeks 1 day ago. Offline
Joined: 08/05/2010
Groups: Void Crushers

can we please fix or roll back main to the version when wiggle meant wiggle, there was no silly band slider, and the protein didn't turn into a rock after two rebuilds.

Joined: 09/21/2011
Groups: Void Crushers

One of my group just tested set WA and get WA. set WA does not change WA, get WA returns allways 3. Maybe changing the default to lower would get the old behavior back. The stiffning effect is really affecting my game.

Joined: 06/17/2010

LWS is affected too.
Just started walking my 516 - getting only fractions... I was NOT use any LWS script on this one. It was newer happen b4.

Joined: 11/10/2007
Groups: Window Group
Status: Open » Open

Here are some comments:

Rav3n_pl: Thanks for the video. When I build my own version from 9/28 (before any new sliders were added) and run the script, I get the same behavior shown in the video: extremely short wiggles that get fractions of a point. So from this particular example it seems this is not an old client/new client issue.

auntdeen: The currently released version does also include some bug fixes with wiggle that we found over the past two weeks. This could account for the different numbers, and I believe these could make it run a little longer on each iteration. However, running Rav's script from a reset of the Anthrax puzzle, I get considerably different numbers than you posted. I'll post those shortly.

Timo: We left the "behavior" scripting functions in place, so that scripts wouldn't break, but they don't do anything. But you are correct, I just double-checked the "internal" defaults of WAcc and it seems they are still different from what they were originally, we somehow missed that when reverting them. We'll fix that and get an update posted.

Joined: 11/10/2007
Groups: Window Group

Here is data from my run of rav's script on multiple versions. "Current" is the currently released version of the game (buildid: 2012-02-14-47446:47447), "Old" is my build of a pre-sliders version (would have a buildid like: 2011-09-28-45007). Note the first three numbers in the buildid are the date. I ran on both versions with the rav's default script and with a change to do_global_wiggle_all(3). Here's what I came up with, from resetting the Anthrax puzzle and running the script:


CI=0.01 Score: 8870.8569758962
CI=0.02 Score: 8870.8569675682
CI=0.03 Score: 8870.8569592317
CI=0.04 Score: 8870.856950887
CI=0.05 Score: 8870.8569425343
CI=0.06 Score: 8870.8569341724
CI=0.07 Score: 8870.8569257968
CI=0.08 Score: 8870.8569174085
CI=0.09 Score: 8870.8569090104
CI=0.1 Score: 8870.8569006037


CI=0.01 Score: 1084.1830860715
CI=0.02 Score: 5342.0007854218
CI=0.03 Score: 6645.7260556348
CI=0.04 Score: 7129.8377437851
CI=0.05 Score: 7942.1081038156
CI=0.06 Score: 7946.3144831861
CI=0.07 Score: 8080.9864318513
CI=0.08 Score: 8207.5435366384
CI=0.09 Score: 8870.8726879444
CI=0.1 Score: 8870.8722712436


CI=0.01 Score: 8870.8569750271
CI=0.02 Score: 8870.8569666993
CI=0.03 Score: 8870.8569583626
CI=0.04 Score: 8870.8569500179
CI=0.05 Score: 8870.8569416651
CI=0.06 Score: 8870.8569333035
CI=0.07 Score: 8870.8569249279
CI=0.08 Score: 8870.8569165392
CI=0.09 Score: 8870.8569081414
CI=0.1 Score: 8870.8568997347


CI=0.01 Score: 8870.8578295709
CI=0.02 Score: 8870.8578714052
CI=0.03 Score: 8870.8578796847
CI=0.04 Score: 8870.7138208444
CI=0.05 Score: 8870.7153645254
CI=0.06 Score: 8870.7169047139
CI=0.07 Score: 8870.7184420457
CI=0.08 Score: 8870.7199766348
CI=0.09 Score: 8870.7215085633
CI=0.1 Score: 8870.72

Keeping in mind this is just one structure, some observations:

- CI needs to be relatively low (<0.1) to have an effect. Perhaps we should set up a nonlinear scaling, so that what is curently CI 0.1 would become CI 0.5, for example. This would provide more precision over "useful" low values of CI.

- do_global_wiggle_all needs to be called with a higher number of iterations to see an effect.

- it appears the current version of the client is the most "variable" in terms of response to CI.

bareego's picture
User offline. Last seen 1 year 44 weeks ago. Offline
Joined: 09/18/2010

It would be good if it could be adjusted in a way that older scripts work the same way they used to.
To me it seems that this does not only relate to the Clash but also the way that wiggle all is implemented.
I much preferred the way it was in much older clients (eg 2 years ago).

Joined: 11/10/2007
Groups: Window Group
Status: Open » Closed

This fix has been implemented and is now in the developer preview version.

Joined: 11/10/2007
Groups: Window Group
Status: Closed » Closed

This fix has gone out in the most recent update.


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