Best RMS ??? (Please look Howard)

[Insidious]

Originally, I was only going to send an e-mail, but changed my mind because I am just toooooo curious.

Why is it that my Best RMS can stay constant for days and days, and then when I delete my distribfold directory and re-install DF, I get a new (and better) Best RMS?

I have seen this on 3 occasions now. The first two I pretty much ignored, but this last time it was on purpose.

I suspect there is something amiss! (might want to give it a look)

-Sid

[FoBoT]

this thread (http://www.free-dc.org/forum/showthread.php?s=&threadid=1927) discusses the randomness that should drive anyone finding a lower RMS structure

maybe you got a lucky box?

[Insidious]

Yeah, I saw that a while back.

It describes how everything is supposed to work.

I am suggesting that something may not be working as expected or desired.

I'm not sure how the Best RMS value for a user is stored, but I have begun to wonder if there may be an updating issue related to that.

I'm thinking that RMS is the goal of all our crunches, it is an area that Howard would want to investigate if anything seemed irregular.

PS: Why are we typing instead of eating? :)

[m0ti]

I think that I felt that happen too a couple of times.

As in it would stay constant for days and then I'd install on a new machine and the next day it would drop.

Not that I did any hard checking or anything, but I agree; there's definitely a feel that the best scoring structures tend to be outputted in the first couple of days for a protein and then it take a LONG time to find anything better. If it was TRULY random then one would expect to find a better structure in 4 days of folding time than in 2 days. Even though it's only approximately random, I don't believe there's any chance correlation between the algorithm chosen and the distribution of structures (the chances of that would probably be about 1 in a google or something).

[Brian the Fist]

We have shown that the RMSDs are distributed according to an EVD (Extreme Value Distirbution) and so you can use this to compute the chance of getting a 'better' RMSD that a particular value - it gets smaller and smaller pretty fast as the RMSD threshhold goes down. Installing on a new machine cannot have any effect on your best RMSD other than by chance

[FoBoT]

i think Insidious has one of these in his case ;)

http://www.policefcu.org/mar02/Four%20Leaf%20Clover%2014.jpg

[Insidious]

GIVE THAT BACK!

(Thanks Howard)

[Mikus]

Originally posted by Brian the Fist
We have shown that the RMSDs are distributed according to an EVD (Extreme Value Distirbution) and so you can use this to compute the chance of getting a 'better' RMSD that a particular value - it gets smaller and smaller pretty fast as the RMSD threshhold goes down. Installing on a new machine cannot have any effect on your best RMSD other than by chance
The original question referred to a situation where the RMSD value set early in a "period" was not being improved upon despite a long subsequent "run". Wouldn't the probability of getting THAT particular RMSD value remain constant for the user, rather than getting smaller and smaller? If so, was it only by chance that THAT value was recorded early on, instead of later on?

And when, after a "long dry spell", an user happens to begin a new "period" (such as by reloading the client and thereby changing the "header content" of the uploaded filesets) and thereupon finds his RMSD value bettered, the question is again: Was it only by chance that the better RMSD value was recorded so early in the new "period"?

mikus

[Insidious]

>cannot have any effect on your best RMSD other than by chance<

It wasn't really the calculational algorithm I was questioning, it was the handling of the result.

Is it possible that the project has some sort of difficulty updating the best RMS under some unforseen circumstance (such as it is not the first 'best' found?

[Brian the Fist]

I cannot listen to any of this vague postulation. If you think this is a real problem, I need facts and a controlled experiment. If you are convinced you are right, record your best RMSD, how many structures you make, and then how long it takes after you restart til you get a better RMSD, and record the new RMSD as well. If you can provide me with solid numbers like this I can better tell you how likely it is that it is expected behaviour.
I am currently of the opinion that it is just paranoia however :D

[Insidious]

Okies,

Really wasn't trying to attack the project or anything.. I was just trying to give input and point out a vauge (potential) wierdness. I totally respect your judgment as to whether or not it is anything to be concerned about.

Thanks.

[Paratima]

Originally posted by Brian the Fist
I am currently of the opinion that it is just paranoia however :D Diagnosis of paranoia does not necessarily preclude the fact that someone or something, somewhere, really IS out to get you! :eek: