In playing around with an IRLS operation I noticed that
calling lsfit with a weight vector containing no zeros leads to
the following error:
> lsfit(rnorm(100),rnorm(100),abs(rnorm(100)))
Error in xzero %*% z$coefficients : non-conformable arguments
By contrast, if there are some zero weights, no errors are generated:
> lsfit(rnorm(100),rnorm(100),round(abs(rnorm(100)),0))$coef
Intercept X
-0.1007470 0.1072742
The problem is with the following part of the function:
if (!is.null(wt)) {
if (ncx == 1)
fitted.zeros <- xzero * z$coefficients
else fitted.zeros <- xzero %*% z$coefficients
z$residuals[wt == 0, ] <- yzero - fitted.zeros
z$residuals <- z$residuals * invmult
}
This only works if NROW(xzero) >= 1, where xzero is defined as
xzero <- as.matrix(x)[wt == 0, ] .
An easy workaround is to use
if (!is.null(wt)) {
if (NROW(xzero) >= 1) {
if (ncx == 1)
fitted.zeros <- xzero * z$coefficients
else fitted.zeros <- xzero %*% z$coefficients
z$residuals[wt == 0, ] <- yzero - fitted.zeros
}
z$residuals <- z$residuals * invmult
}
--
Stephen Shiboski <steve at biostat.ucsf.edu>
Division of Biostatistics
University of California, San Francisco
=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-r-help
mailing list -- Read http://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html
Send "info", "help", or "[un]subscribe"
(in the "body", not the subject !) To: r-help-request at
stat.math.ethz.ch
=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=