This is the home page for the dataspork space.
Download current version at MCC Software archive
Subversion download:
Background, manual on dataspork
http://www.mcc.uiuc.edu/dataspork/
Note that no installation step is necessary with the pre-built jar.
This applet (signed by Jeongnim Kim, UIUC) can be used to analyze the data files on your desktop. In order to use it, check if your browser allows an applet to run.
Start dataspork
If the applet does not start:
Ideas for DataSpork enhancements
Let us discuss here enhancements to DataSpork.
Better error analysis (DMC)
I recently read the paper by Wolff on hep-lat/0306017 (Comput.Phys.Commun. 156 (2004) 143-153) and I have some reactions/ideas that could impact future dataspork.
- We should add the error of the error (e of e) , discussed there. This should be done both in the trace (as a band of errors) and in the numerical output for example average = -2.0 average error = .21 error range =( .185, .225).. Wolff gives an explicit formula Eq. (40) which can be used. It involves the window (more below).
- Wolff suggests that the acf method is more accurate than the bining method (fig. 1). I haven't seen this before. Perhaps this should be apparent in the dataspork output of the e of e.
- Wolff has a method for selecting the window (upper limit for the acf integration). As far as I can see this is based on the fact that the acf decays exponentially. (eq 36). This may be true, or roughly correct, but it would be nice to a) have some data and b) have a method which is insensitive to this assumption and c) fit the acf to an exponential within dataspork and show it on the acf figure. Once you fit the acf to an exponential, then he picks W to minimize the sum of the neglected tail of the acf and the eofe (Eq 43). Do we want to compare to whatever Shumway picks W to be in the current algorithm and give the user an option of the window?
- on page 6-7, he discusses an analysis method for replicas, to see if they are consistent. Perhaps a start to what we talked about in the endstation. An automatic way to decide if data from different runs is consistent.
- I could not find any relevant citations to this article--it would be nice to know if there is more discussion about estimating errors, in such a way as to minimize the eofe. Note that this has impact on the penalty method and on petascale computing, since we want to do all of this in a precise, robust fashion.
