Folding Forum

I wonder if anyone from the Pande Group fold as well? And what kind of hardware are they running? And what are their stats like?

Yes, Pande staff do fold. All kinds of hardware. Do a stats search on names like "Pande" to find their team and team members. (Team #1 )

Search is your friend.

Yea...I didn't know what the search string would be.

Do they count their development WUs in with their own stats or do they fold just like everybody else does?

(How in the world does anonymous have 70M+ points? - rhetorical question)

alpha754293 wrote:Do they count their development WUs in with their own stats or do they fold just like everybody else does?

I'll bet that most development WUs don't earn any points. What you see is the extra work done on the unused resources, not the development setups.

codysluder wrote:

alpha754293 wrote:Do they count their development WUs in with their own stats or do they fold just like everybody else does?

I'll bet that most development WUs don't earn any points. What you see is the extra work done on the unused resources, not the development setups.

Well, if WUs that have problems, errors, bugs, etc. - those that we don't get credit for would be what I'd think fall into the development group.

I would think that if they develop the next group of WUs based on the results from the previous group that they'd have to develop them. (they don't just magically appear). And that they'd have to work on them so that they'd be stable enough for the general public to run.

I think Cody's point is that there are developmental WUs which are processed by the Pande Group before they're benchmarked. The points are initially set to some meaningless value (say 10 points) to be replaced by a realistic value when they're benchmarked. Even if we assume there are no EUEs, for those WUs, the points can't be used as a measurement of their participation.

Are all WUs benchmarked before release then? Or do they just do statistical sampling?

alpha754293 wrote:Are all WUs benchmarked before release then? Or do they just do statistical sampling?

Either that's a nonsensical question or I don't understand what you're asking.

If the Pande Group measures the time to complete every WU, they don't need us . . . . all of the calculations will have already been completed.

codysluder wrote:I think Cody's point is that there are developmental WUs which are processed by the Pande Group before they're benchmarked.

If you're working in a production environment (say a factory for example) that make 100,000 parts a year, you're not likely going to run statistical analysis on all 100,000 parts to see whether they're in compliance with the specifications. Therefore; there's this little thing called statistical sample where out of each production batch/run, a number, x, is selected at random and analyzed as a representative sample of the entire population. Based on that statistical analysis, it the is used to extrapolate information about whether or not the rest of the population is also within specification or not.

Therefore, according to the same production engineering statistical theory, if Pande Group (for example) generates 100,000 WU's per year, do they benchmark each and every single WU or do they benchmark a select few, that are more presentative of the projects and/or project groupings (where applicable)?

(And I don't believe that they would have to run the entire WU in order to determine the benchmarked value, which renders your last point irrelevant. I mean, why would you need to run the whole thing to do the benchmarking?)

AFAIK, they benchmark every project -- probably a portion of one WU, but that's only a guess. There is no statistical sampling.

Benchmarking is required to keep the donors happy, but it has no scientific benefit, and their objective is to do a maximum amount of science.

I though we were talking about developmental WUs and now it seems the subject has changed to benchmarking. Those are NOT the same things.

But wouldn't benchmarking EVERY project be quite time consuming?

Seems to me that would be completely counterintuitive/counterproductive (in relation to your closing statement).

Certainly if I'm going to study protein XX at three different temperatures, we could debate whether I'd need one or three benchmarks when I release those three projects. If after a few weeks I make minor adjustments to them and create three new projects, I might not run new benchmarks unless I changed something significant.

We've had several thousand new projects over ~10 years, so that might be a few benchmarks per week distributed across several project owners.

New projects often begin testing with an announcement something like this. You figure out if new benchmarks would be performed.

xxxx wrote:Project xxx6 is due for release shortly, to be followed by xxx7. Both are . . . similar to xxx0-xxx5. We are expanding our studies of xxxxxxxxxxx in this series. WU's from both projects are worth xxx points. Preferred and final deadlines are . . . . .

bruce wrote:Certainly if I'm going to study protein XX at three different temperatures, we could debate whether I'd need one or three benchmarks when I release those three projects. If after a few weeks I make minor adjustments to them and create three new projects, I might not run new benchmarks unless I changed something significant.

We've had several thousand new projects over ~10 years, so that might be a few benchmarks per week distributed across several project owners.

But see..if you're developing the WU though, right, in the early stages of development, if and when you do get to the point where you actually start benchmarking it, you might progressively benchmark it at various points in time.

So, unless you hold all benchmarking until just prior to release so that you'd only ever do it once.

On the other hand, you can find out that when you benchmark it, that as it's running, you run into problems, so you end up back at the drawing board, which then, over time, causes to to benchmark periodically.

I guess that it depends on how the WUs are developed.

For the stuff that I do at work, we are required to do verification and validation checks every so often. My boss does the development V&Vs and I do the production large-scale V&Vs. Both of us keep tabs on the performance of the code/program as well for job scheduling (especially in my case cuz when you're doing 900 runs where each run takes 3 weeks). Eventually you get so used to it that you can dynamically benchmark each run as it's running while starting the next batch.

I think we're talking about two different things. What I call development is figuring out how to arrange the protein (and solvent, if that's applicable) so that the analysis will run properly -- generating the input data -- confirming that they represent the problem that is to be studied -- that the best choice of analysis method has been made, etc.

In the example mentioned in my earlier post, there probably was considerable development prior to releaseing project xxx0. Projects xxx1 through xxx7 probably wouldn't have little or no "developement" -- they'd go straight to beta testing. If beta testing turns up problems, the developmental steps can be rechecked and projects xxx8 and xxx9 created. Projects xxx7 and xxx8 might never reach public release.

Do you consider the WUs that we do public or beta (like the clients)?

If it's public, then they have to have benchmarked at least some projects or parts of all projects.

If they're betas, then they still need to benchmark them (at least in part) in order to determine the point value (or they might just plug it into some formula that they've got sans benchmarking), assign it, release it, and have us do the "benchmarking" for them (or run it) or whatever it is that you want to refer to as benchmarking.

As you've mentioned several times, I think that they've got more important things to do than benchmarking, so it would seem highly unlikely that they would benchmark all projects, even ones in beta/development, and certainly not the public release ones.

If I were the one doing it, I'd probably just come up with some kind of quickie math equation/formula and not bother spending any more time than that (than I have to/is required), so that I'd be able to focus on "the science".