Results 1 to 17 of 17

Thread: Interesting observation...

  1. #1
    Boinc'ing away
    Join Date
    Aug 2002
    Location
    London, UK
    Posts
    982

    Interesting observation...

    I have a Dual MP2400+ running Windows XP and Folding...I have the latest client...and I have both clients turning out energy values of 41.776 for gen 0 for the past 3 sets they have done...yep - both clients turning out 41.776 for gen 0 for each of the 3 sets done so 6 lots of 41.776 at gen 0...

    Just thought it was an interesting observation

  2. #2
    That does sound odd though not impossible. If you continue to find this or if gen 1, 2, 3 etc are also all the same let us know. What about the 'best energy', was that all the same too? What was it? Was 41.7 the value for 'RMSD' or 'Energy' and where are you looking exactly?
    Howard Feldman

  3. #3
    Sneaker pimp
    Join Date
    May 2003
    Location
    In-gur-lund
    Posts
    24
    I've had all 4 of my machines do that on gen # 0

  4. #4
    Boinc'ing away
    Join Date
    Aug 2002
    Location
    London, UK
    Posts
    982
    Originally posted by Brian the Fist
    That does sound odd though not impossible. If you continue to find this or if gen 1, 2, 3 etc are also all the same let us know. What about the 'best energy', was that all the same too? What was it? Was 41.7 the value for 'RMSD' or 'Energy' and where are you looking exactly?
    Best energy after a complete set was different...the 41.776 was the best energy for gen 0 - dfGUI, DCMonitor and progress.txt all reported it...

    I'll keep any eye on it but it seemed interesting (to me) that both CPUs in a box were getting the same best energy repeatedly...but as you say, it's not impossible

  5. #5
    Boinc'ing away
    Join Date
    Aug 2002
    Location
    London, UK
    Posts
    982
    well, one of the duallies has now gotten 39.990 on it's 5th set - so obviously just a major coincidence...

  6. #6
    Actually I suspect I know why this is occurring, it is an artefact of the way we are computing the generation zero 'fitness score'. It probably means the fitness is being based solely on compactness and how closely secondary structure matches predicted SS. Crease energy is outside of the 'expected' range and so is contributing zero.
    Howard Feldman

  7. #7
    Also, gen 0 takes a very long time to caclulate, perhaps even longer than the previous protien. Considering how quickly you get through a set of 250 gens. for this protein, it seems a little strange.

  8. #8
    Have two clients installed. Run one off-line until you have close to 250 generations in the bank.

    Start to upload the cached gens. This will take a while but the client did not have to stop to upload every generation beforehand.

    While the upload is going, start the other client and run through until near the end.

    Repeat.

  9. #9
    Originally posted by HaloJones
    Have two clients installed. Run one off-line until you have close to 250 generations in the bank.

    Start to upload the cached gens. This will take a while but the client did not have to stop to upload every generation beforehand.

    While the upload is going, start the other client and run through until near the end.

    Repeat.
    Makes no difference if you average it over a week, you still have to caclulate the 250 generations offline which will slowdown the client that is online by about the same amount.

    I don't actually understand the function of generation 0, nor do I know why it has 10,000 structures, all I know is that it chews up a lot of kilowatt hours over a week that I have to pay for out of my own pocket, so I hope it's usefull.

  10. #10
    Vorlon Ambassador to F-DC Kosh's Avatar
    Join Date
    Mar 2002
    Location
    Toronto, Ontario, Canada
    Posts
    88
    Well from what I understand generation zero is the starting point, you make 10000 completely random proteins. Each generation after that takes 50 proteins from the previous generation and improves them. Starting with too small of a base would greatly reduce the probablity of finding reasonable structures. So your kilowatt-hours are going to good use -- as far as I know anyways.


  11. #11
    Originally posted by Kosh
    Well from what I understand generation zero is the starting point, you make 10000 completely random proteins. Each generation after that takes 50 proteins from the previous generation and improves them. Starting with too small of a base would greatly reduce the probablity of finding reasonable structures. So your kilowatt-hours are going to good use -- as far as I know anyways.

    If it's important then generation 0 should be worth up to 20,000 points as it takes about 15% of the time/energy required to complete a set, yet at the moment it only contributes .037% towards the score for a set, yet the cpu works just as hard doing generation 0. I was under the impression that the scoring system was suppose to be some sort of a representation of the caclulating resource effort put in by the teams/individuals, but it seems to be arbitrary.

  12. #12
    Fixer of Broken Things FoBoT's Avatar
    Join Date
    Dec 2001
    Location
    Holden MO
    Posts
    2,137
    Originally posted by erk
    I was under the impression that the scoring system was suppose to be some sort of a representation of the caclulating resource effort put in by the teams/individuals, but it seems to be arbitrary.
    this scoring system was designed to reward those that run all 250 generations, as the later generations generally produce better "Science" (results)

    it is an inducement to run the program "normally" vs. just restarting gen 0 to rack up points
    Use the right tool for the right job!

  13. #13
    OCworkbench Stats Ho
    Join Date
    Jan 2003
    Posts
    519
    If you are worried about your slow PCs taking forever on Gen 0, get a fast computer to do Gen 0 then xfer the Directory to the Slow PC. Crazy yes, but it worked for my Duron 750. For more dumb suggestions just ask
    I am not a Stats Ho, it is just more satisfying to see that my numbers are better than yours.

  14. #14
    Originally posted by Grumpy
    If you are worried about your slow PCs taking forever on Gen 0, get a fast computer to do Gen 0 then xfer the Directory to the Slow PC. Crazy yes, but it worked for my Duron 750. For more dumb suggestions just ask
    All my computers are fast so that's not the point, the original poster observerd that the energy value of generation 0 was ver high which made me query the value of generation 0.

    My logic is:

    If generation 0 is valuable then give it more than 50 points.

    If generation 0 is not valuable then get through it faster and on to the serious stuff.

    I think the argument that generation 0 has low points to encourage people to continue on to later generations is flawed. How many people stop at one set? Bugger all I would say, most folk keep going on and on producing sets, blowing at least 15% of their time stuck on generation 0 for little reward either in terms of a low energy result or points score. I would like to see that time reduced to well less than 10%.

  15. #15
    Originally posted by erk
    If generation 0 is not valuable then get through it faster and on to the serious stuff.
    The problem is, we cant go faster through it ... a completely random computation is the fastest way to generate those structures we need to go through the other 250 generations.

    Atm, our project goal is to optimize the algorithm, therefore we need to go through the whole 250 generations ( what we do at generation 0 doesn´t distinguoish much from Phase I what we have done enough )
    The German DC Community : Team Rechenkraft.net - Join now ! Rechenkraft.net

  16. #16
    I have run 18 complete 250's on my main rig, and everyone started with 39.990 for the first generation.

    Saying that, I've had everything ranging from 3.26 up to 7.81 for each completed 250, so it's obviously not a problem.

  17. #17
    Junior Member
    Join Date
    Apr 2002
    Location
    Toronto, Canada
    Posts
    27
    Originally posted by erk

    My logic is:

    If generation 0 is valuable then give it more than 50 points.

    If generation 0 is not valuable then get through it faster and on to the serious stuff.

    The problem is that gen0 is not valuable in the structures it generates, but absolutely necessary since it provides the foundation for gens 1-250. Since it is not valuable, it only has a point value of 50. Since it is necessary, we generate 10,000 structures.
    Derek

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •