--- [EMAIL PROTECTED] wrote: > Here's a little snippet on linear regression: > http://www.mathworks.com/access/helpdesk/help/toolbox/stats/linear20.shtml#49537
Nice reference. Unfortunately, it points out the fact that I think either you, J or Al mentioned some time ago that solving the normal equations directly gives bad numerics. That means that it is probably *not* a good idea to use RealMatrix directly, but maybe to first implement QR decomp. In any case, we need to look into the numerics, select a good algorithm and probably make it pluggable. > > The first equation is the common form (in matrix terms) of a linear > model. It also defines residuals and other important points regarding > multiple regression. > > The nice thing about linear models, is not only are they used for > regression analysis, but they are the basis for ANOVA as well. So, I > for one, would like to see a pure matrix based implementation for > linear regression as it more likely reuseable for ANOVA. > > Also, we can add a facade to the matrix implementation for those whole > just want to deal with double[] and double[][] structures. Much like > we've done with the summary stats. +1 > > Take note of the hat matrix. Some more advanced analysis like > measuring the effect of an observation can be easily accomplished > using the hat matrix. So, it might be worth while to incorporate > caching of the hat matrix in any implementation. +1 Phil __________________________________ Do you Yahoo!? Yahoo! SiteBuilder - Free web site building tool. Try it! http://webhosting.yahoo.com/ps/sb/ --------------------------------------------------------------------- To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]