Hello, Unless someone wants to buy the West's reference, the closest thing to an explanation for what is in those classes is in the Sandia report.
Perhaps you can add a link to the Sandia reference with the qualification that it is not an exact description of the algo? Cheers, -Ajo. On Sun, Nov 17, 2013 at 8:10 AM, Phil Steitz <phil.ste...@gmail.com> wrote: > On 11/17/13 6:02 AM, Ajo Fod wrote: > > The Sandia report is free. The javadoc link is paywalled. I'd recommend > > linking a free version if you want people to have unfettered access. > > > > Does the paywalled document describe how to do the > aggregate/combine/reduce > > operation we talked about in > https://issues.apache.org/jira/browse/MATH-1016? > > The Sandia report does. > > > > Also, I've been working off the Sandia report and I think that adequately > > describes what is going on in the classes. Let me know if you are aware > of > > any differences. > > What the moment classes do is described exactly in the class javadoc > inline. Per comment above, SecondMoment is *not* what the Sandia > reference describes. West's algorithm is *not* what that reference > describes either. If you can find a free reference to West's > algorithm for the variance computation I will gladly add it. > > Phil > > > > Cheers, > > -Ajo. > > > > > > On Sat, Nov 16, 2013 at 11:34 AM, Phil Steitz <phil.ste...@gmail.com> > wrote: > > > >> On 11/16/13 3:55 AM, Ajo Fod wrote: > >>> Hello, > >>> > >>> The document: > >>> > >> > http://commons.apache.org/proper/commons-math/javadocs/api-3.2/index.htmlis > >>> relevant to First/Second/Moments, could someone add it to those > >>> classes > >>> as well? > >> The javadoc for first and second moments and variance now describes > >> exactly what the updating algorithms are. The moments use > >> recurrences similar to what the reference you provided suggests, but > >> they are explicitly stated in the javadoc and pretty elementary. In > >> general, when we can provide full algorithm descriptions in [math] > >> javadoc, we prefer to do that, as it makes it clear exactly what we > >> are computing. If what you are suggesting is that we change the > >> updating formulas, please clarify and explain the numerical > >> rationale for the change. The (storeless) variance computation uses > >> West's updating algorithm [1], as documented in the class javadoc > >> for Variance. One other thing to bear in mind here is that what we > >> call "SecondMoment" is a not really the standard second moment > >> statistic. The difference is made clear in the class javadoc. This > >> statistic is a building block used by other stats and its definition > >> reflects that. > >> > >> Phil > >> > >> [1] http://doi.acm.org/10.1145/359146.359152 > >>> > >>> This link is already documented in : > >>> > >>> > >>> > >>> > >>> *org.apache.commons.math3.stat.correlation.StorelessCovariance* > >>> > >> > >> --------------------------------------------------------------------- > >> To unsubscribe, e-mail: dev-unsubscr...@commons.apache.org > >> For additional commands, e-mail: dev-h...@commons.apache.org > >> > >> > > > --------------------------------------------------------------------- > To unsubscribe, e-mail: dev-unsubscr...@commons.apache.org > For additional commands, e-mail: dev-h...@commons.apache.org > >
