>In a previous post, I had asked how matrix accum works. Nick Cox
>replied that it was a very clever algorithm proprietary to Stata.
>Stata's awesome :)
>
>Maybe I should rephrase my question. How might one go about calculating
>such things as cross-product matrices without swamping memory by storing
>entire lists of independent variables as matrices?
Why not use -matrix accum- for that?
If you are asking to do it by hand, then you need to take the sum of
the product of two variables (or with itself or with ones). You can
use -sum( )- function for that. It comes with an exceptionally high
degree of precision, which leads me to believe this is one of the
purposes for which this ugly-duckling was designed for. Most people
don't use it much, briefly becoming undocumented when a
similar-sounding option became popular.
My guess is that it or its cousin is responsible for making fast and
precise computations of OLS-type estimators possible. Wrapping this
function into a null mat, you should be able to reproduce the results
of -matrix accum- almost as fast. It produces a running sum, so you
will need to collect the results from the bottom row.
You can alternatively take the product of two variables and use -sum-
command to get the mean and the number of observation. Multiply them
together, and you again get the sum of the product of two variables.
Roy
_________________________________________________________________
Windows Live™ Hotmail®: Chat. Store. Share. Do more with mail.
http://windowslive.com/howitworks?ocid=TXT_TAGLM_WL_t1_hm_justgotbetter_howitworks_012009
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
