[S] loess SE estimate runs out of memory

Purvis Bedenbaugh (purvis@phy.ucsf.edu)
Wed, 15 Jul 1998 22:32:42 -0700


HI All,

These days I'm making summary figures based on many
data points, and many programs are failing for lack of memory.
Often the analysis can be demposed into sub-sections, but this
one has me stumped.

The job is to estimate the standard error bars for group data from
several experiments. Even reducing the request to estimating standard
error at one data point, this fails on a DEC Alpha/AXP with 256
megabytes
of memory, running Digital Unix.

> my.fit <- loess(yvals~xvals,data=na.omit(my.data),span=1/2,degree=1)
> my.se <- predict(my.fit, newdata=6,se.fit=T)
Error in .C("loess_ise",: Unable to obtain requested dynamic memory
(this
request is for 153190304 bytes, 158883856 bytes already in use)
> dim(my.data)
[1] 6379 2
>

Is there a more memory efficient way to get to standard error estimates
for an emperically defined curve ?

It's no problem to estimate SE at 15 x values in one call with the
single
experiment data sets, even on a 128 megabyte machine. There are a
total of four experiments in the composite table.

--
Purvis Bedenbaugh
purvis@phy.ucsf.edu
-----------------------------------------------------------------------
This message was distributed by s-news@wubios.wustl.edu.  To unsubscribe
send e-mail to s-news-request@wubios.wustl.edu with the BODY of the
message:  unsubscribe s-news