LWS strategy is not working anymore

Case number:845813-992192
Topic:Game: Tools
Opened by:Rav3n_pl
Status:Open
Type:Bug
Opened on:Tuesday, March 20, 2012 - 22:51
Last modified:Thursday, July 26, 2012 - 17:07

We all found that local wiggle sequence is not working anymore.
If we make totally "fresh" alignment form scratch mad pump it using bands, rebuilds etc it was common that any LWS script make from few to even few hundred points. Now series of them make only 1-2 points!
LWS endgame strategy always make cool score boost and last day/hour/minutes was always exciting because sometimes someone form #20 (or lower!) get in top 5 only by LWS.
We miss that days, please change it back to that state. Looks like changes in wiglle behavior are much deeper than we all thought.
At this point walking it total waste of time.

(Tue, 03/20/2012 - 22:51  |  22 comments)


spmm's picture
User offline. Last seen 32 weeks 4 days ago. Offline
Joined: 08/05/2010
Groups: Void Crushers

yes I agree rav - even in main, which is supposed to be unchanged, LWS is not working, which I suggest means that we are all working with much stiffer puzzles. Also the multiple alignment puzzles often had slightly stretched or compressed segments which shook out during LWS gaining points without stiffening, there seems to be no way to get those out now.
I am also seeing segments which are not glycine or proline bent in half with bad scores (not on hairpins)and they don't rebuild or wiggle out - isoleucine is one in particular, that may or may not be related .

beta_helix's picture
User offline. Last seen 9 hours 55 min ago. Offline
Joined: 05/09/2008
Groups: None
Status: Open » Open

Could this just be an issue with CASP ROLL puzzles?

We have reverted everything back to what Foldit was like before CASP ROLL started, so perhaps it is something about these CASP ROLL targets.

Since we haven't been able to post any other prediction puzzles, it could be something to do with the starting server models, although we are giving you models from 2 very different servers (Quark/Zhang & RosettaServer).

Is this also true with the design puzzles where you are able to design certain regions from scratch?
What about the Beginner Puzzles (since the current puzzle was posted a couple months ago)?

Joined: 06/17/2010

Same thing is in last QTTN puzzles. This is not CASP puzzles only.
I made denovo (not using threading etc.) to about 90% of my best score and tried to walk it - got 2 points...

Joined: 09/21/2011
Groups: Void Crushers

Beta, last night (for me night) it was mentioned in chat that there are some differences, bugfixes.
Could it be that fixes in global wiggle make it so that there is less leftover for local wiggle to fix?

Joined: 12/27/2010
Groups: None

I have noticed for quite a while that walking was not helping me any more but I assumed it was because the wiggles and shakes have squeezed more points out than previously and there is less left for walkers to take. In that case, we haven't really lost anything, just the rush of a final points gain.

I also feel that the competition (not the protein) has become a lot stiffer. I am lucky to make the top 100 folders in a puzzle now whereas a few months ago I was easily in the top 50 or less.

spmm's picture
User offline. Last seen 32 weeks 4 days ago. Offline
Joined: 08/05/2010
Groups: Void Crushers

I agree that the competition has become more intense which is great - more good folders - but I don't understand why walking has changed - walking early often locked your protein into really nasty minima so rather like running with Wiggle Accuracy set high, you get early points but loose flexibility when you really need it.

Joined: 06/17/2010

Yes, only rebuild or relly stong banding was need to make it flexible again. Now it is looking like "pre-walked" every wiggle and nothing can make it flexible again. Maybe "something" is changed in CI http://fold.it/portal/node/990990 ? It really looks like CI is working in "narrow" range.

Joined: 06/24/2008
Groups: Void Crushers

I also have found the flipping side chains is not giving up as many points as previously; I assumed it like walking had lost its effectiveness because of all the play with CI&shaking&wiggling using scripts. Now I wonder if something else has happened.

Joined: 06/17/2010

True. Acid Tweaker and clones not making much on good fuzed one.

itskimo's picture
User offline. Last seen 7 years 3 weeks ago. Offline
Joined: 08/10/2010
Groups: foldeRNA

Yes i have seen the lessening of the points from walkers at the end game, this may have something to do with the casp puzzles because they are generations of the same puzzles, worked on by many many people and every script which include walkers in each genneration of the casp contests?

Like shapening a pencil to a point, we are comming closer and closer to the refind solutions of casp?

My feeling are that the banders and rebuilders may have gotton much better at shaking out those last few points befor you need to walk at the end.

Also, more people may be using more walkers earlier in the game, when they evo from a red line. This distrubuted walking may help to settle the protein sooner and better, thereby reducing the amount of walking that is needed at the end?

Joined: 06/17/2010

Read my comment. I made DENOVO from flatline. I also was not walkable. "Distributed walking" is not the cause. It looks like wiggle after last fix in code is doing much more than b4.

beta_helix's picture
User offline. Last seen 9 hours 55 min ago. Offline
Joined: 05/09/2008
Groups: None

We may have found what is causing this problem, but it is taking a while to pin down.

Hopefully this will finally be resolved soon, thanks again for reporting this and especially for your patience!

Tlaloc's picture
User offline. Last seen 2 weeks 23 hours ago. Offline
Joined: 08/04/2008
Groups: Mojo Risin'

Well, I thought I might have some insight into this, but I don't. I did fix a bug in my RepeatSettle script for Lua V2, but that doesn't fix the underlying problem in local wiggle.

Joined: 06/17/2010

A bump. Is there solution on horizon? We have one good technique less to use.
If it is related to "wiggle fix" maybe release wiggle slider that will use different wiggle standards.
As far i remember some chat fragment wiggle is fixing few things in bbone/sidechain, maybe it is possible to enable/disable that "things" in wiggle?
lvl 1 - lazy/fastest wiggle, even not explode on clash
lvl 2 - better/fast wiggle, jumping out o clash
lvl 3 - old wiggle, we love it
lvl 4 - current wiggle, better but worse...
lvl 5 - wiggle on wiggle accuracy 7 - make rock from any protein

jflat06's picture
User offline. Last seen 5 days 9 hours ago. Offline
Joined: 09/29/2010
Groups: Window Group

The problem is that many of the scripts that you guys have developed are workarounds for some bugs or limitations in the Foldit software. When we fix the bugs or give you more features, some of these scripts and techniques are no longer necessary, or wont work as well.

The most obvious case of this is the wiggle slider (also known as minimization tolerance). Adjusting the minimization tolerance is a very basic ability within Rosetta. It lets you tell wiggle how deep it should dig while wiggling.

However, in Foldit, we always had it set at some particular value. Because of this, players came up with interesting ways to *force* wiggle to keep working when it normally would have stopped.

So what happened when we released the wiggle slider, is that people would wiggle in the early game, and get the protein 'cemented'. Then when late game came around, these scripts which were designed to exploit wiggle into working harder no longer gained points, because it had already been working hard.

When we change the software, our goal is usually to make things more efficient, so that people can come up with higher quality solutions faster. However, when the software changes, the techniques need to change with it.

Joined: 06/17/2010

*trashing all LWS scripts*

Susume's picture
User offline. Last seen 7 hours 5 min ago. Offline
Joined: 10/02/2011

I appreciate the explanation of why LWS scripts don't work any more. I had heard speculation from players but it's nice to hear it "officially."

jflat06's picture
User offline. Last seen 5 days 9 hours ago. Offline
Joined: 09/29/2010
Groups: Window Group

I am not saying that's exactly what happened here. Certainly there still can be bugs which are causing issues.

My only point is that we cant expect all scripts and techniques to keep working as we improve the client.

A similar thing might happen if we were to introduce loop hashing (a tool similar to rebuild). A lot of rebuild scripts might just not be worth running anymore, because their job would be done better by a new tool.

Joined: 06/17/2010

New rebuild tool? *all ears*

jflat06's picture
User offline. Last seen 5 days 9 hours ago. Offline
Joined: 09/29/2010
Groups: Window Group

Lol, we've discussed adding it. The current opinion seems to be that it is 'too good' right now, though. The problem is that the tools the Baker Lab is coming up with are getting much better, but the score function isn't getting better at the same rate.

So we have these great new tools, but without a proper score function to guide them, they just cause us to get really great scores on any structure, which actually hurts more than it helps.

Joined: 06/17/2010

GIMME DA TOOL! :P
Create beta puzzle that allow this tool (like cut tool limitations) and let us test it :)
Just duplicate any other puzzle we have at the moment to compare results.

jflat06's picture
User offline. Last seen 5 days 9 hours ago. Offline
Joined: 09/29/2010
Groups: Window Group

There's a lot of work involved in getting the tool implemented in Foldit, and we're prioritizing other things right now. If the Baker Lab manages to improve the score function, we'll consider this option more seriously, but it's on the shelf for now.

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