Stats?

[prokaryote]

Hi Miguel,

Just curious as to what kind of equipment team Sussex is running! Wish I had some. :)

Also, please see This Thread (http://www.free-dc.org/forum/showthread.php?s=&threadid=4400) for an example of a stat that may be adaptable for the running of different circuits. A weighting factor similar to what you have described before could easily be implemented.

I know that several people were put off this project because of the discrepancy in circuits analyized running a large circuit vs a small test circuit.

Thanks,

prok

[michaelgarvie]

The Work done in Effort Units is now a fair metric of processing power. It is basically a weighted sum of circuit evaluations weighted on a pretty accurate estimate of the cost of each evaluation.

[michaelgarvie]

btw those stats graphs look really cool. When are you going to set up your own page with cross project stats? Or are there any good ones out there?

[prokaryote]

Hi Miguel,

Cool about the new "points" system! :)

Thanks, I used a very simple Matlab program to generate the values and Excel to chart. I had to manually enter the rawdata though since I haven't learned php, perl or some other scripting language yet (but I will eventually).

Dyyryath is going to look at it over the weekend and see what he thinks (along with feedback from other Free-DC people) so it may be hosted on this site or on Dyyryath's excellent stats page Here (http://stats.zerothelement.com/index.php) or not at all, if no-one is interested.

There's at least two combined stats pages that I know of. One Here (http://www.msu.edu/~bercikda/index.html) that is similar to what is being proposed, but it's "distance measure" warps the data space and exagerates the ranking statistic along the diagonal (0,0,0...0) to (1,1,1...,1) or close to it (since the axis are weighted). The net effect is that it penalizes teams that don't evenly distribute their crunching amongst the various projects. Not an unbiased measure of ranking (independent or regardless of the project weighting that is used).

And One Here (http://www.ninjamicros.com/vbulletin/sss_stats.php) that is also a relative based statistic, but it deals only with rank and not the raw data. Thus it's discrete and you lose information. An example is if in one project someone is rank 1 and say 3X further ahead in points than the rank 2 place. Their given the same super star stat as someone that is rank 1 and only say 1 point ahead of rank 2. That space between ranks information is lost.

There may be others, but I don't know about them. Maybe someone else knows?

Oh yeah, still interested to know what kind of equipment is being used by Team Sussex, seems like they have a couple of super-computers? :notworthy

prok

[michaelgarvie]

DC Rep looks like the most complete one. But it looks like the Free-DC team stats on the DC Repository site aren't being collected properly, what do you think? Maybe its because I've just told Dave to start using the Work Units stat... but some other teams have been getting some score..

Anyway team Sussex is made up mainly of all the Undergraduate lab PCs from the SciTech dept. of Sussex Univ. What was happening was that loads of PCs were submitting under the same name so it looked like one giant one. This has been fixed now..