my observations (dev chat topic on current clients being a little buggy)

Case number:699969-991806
Topic:General
Opened by:karstenw
Status:Open
Type:Bug
Opened on:Thursday, February 9, 2012 - 14:11
Last modified:Monday, February 13, 2012 - 01:19

A summation of my observations

This is not scientific as it relies on my memory and not on good data collection from a control group and an experimental group. These observations may help narrow where the problem is occurring however. It is meant to try and Identify Problem.

1. The rebound: When a script moves from low ci to high ci in its attempt to determine the point value of its newly altered state, it does not wiggle out to as high of a score that it used to. Therefore it rejects its latest altered state more often then it used to. This causes it to score less frequently or not at all.

2. The wiggle all function is not wiggling out to its full potential. There are 2 components to wiggle all: wiggle side chains and wiggle backbone. It appears that this is a problem with the wiggle side chains component (see 3.).

3. Wiggle side chains function is performing in a chaotic manner and often times it "sticks" and does not wiggle out sidechains. At this point it appears that the wiggle backbone function is able to cause an unusually large point increase.

It appears that there are two "areas" of coding that need to be examined. The first area is in sidechains calculations where the "wiggle sidechains" function is used, and an error perhaps caused by the new sliders.

The second area is more difficult to articulate, but could be easily overlooked. This area looks at the setup of the protein itself. It appears that the proteins are behaving in a similar way to folding at low ci techniques even when being played at maximum ci. In this case perhaps the problem would occur regardless of what client is used. Yet different clients would get different results because the protein is responding in an unstable manner to the lower ci.

(Thu, 02/09/2012 - 14:11  |  1 comment)


Hanto's picture
User offline. Last seen 3 weeks 6 days ago. Offline
Joined: 05/10/2008
Groups: None

"Yet different clients would get different results because the protein is responding in an unstable manner to the lower ci."

The above is a quote from last sentence of karstenw's summation of observations.

I don't deny he is seeing what he sees, as I have been using this very same difference in clients results running from the same source solution since I first learned how to play multiple clients on then multi-client folders and now on single client folders. I don't pretend to know what causes it, only that it exists and that I intend to keep on using it to find best available solution on multiple folders using same recipes for same source solutions. I have seen to many times that a single client will not give nearly as much as it should, yet if one broadens the client base to 3 from the same source solution, one can almost expect to get differences, and in many cases one of the resultant solutions will be superior to the other two clients.
I also think most of us, in one way or another, have seen the same as karstenw has mentioned here. However what I perceive in the way the clients are behaving now compared to how they behaved when I first started playing quite frankly is very minor if at all existent.

What I am seeing is an incredibly high number of rotomers using the AT 1.1 recipe which I believe might be related to some recent changes to the client coding and that these increased numbers of rotomers may smooth out the differences between clients that I have learned to depend on in the past. If this is the case, then I will be able to reduce the number of clients I put onto source solutions, which in the end will mean I will be able to play more puzzles. I don't know though, others have better eyes then I do

I do know this tho, AT 1.1 is taking a tremendous amount of time to run because of these increased rotomers, and personally I would rather work with multiple clients running from the same source solution, rather then deal with this much time spent on a single AT.

Variation is the key, to try to keep everyone's client doing exactly what the other person's client does, does not invite the needed variation that leads to evolution and greatly lengthens the time required to work a solution. I also believe this lack of variation will lead to some very boring game play. However I am sure that this lack of variation may prove beneficial to those with the fastest machines capable of persuasively completing the greatest numbers of recipes.

Just my observations over the course of my time here.

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