As was expected, the
runderiv() function has been both useful and deficient. Useful because it offers a good replacement for
smooth.spline() calculations of derivatives for things like N^2. And deficient because it only calculated derivatives, not values!
Both an extension and a renaming were called for. The result is
Below are the examples from its manpage, with the results.
1 2 3 4 5 6 7 8 x <- 1:100 y <- 1 + x/100 + sin(x/5) yn <- y + rnorm(100, sd=0.1) L <- 4 calc <- runlm(x, y, L=L, deriv=0) plot(x, y, type='l', lwd=7, col='gray') points(x, yn, pch=20, col='blue') lines(x, calc, lwd=2, col='red')
1 2 3 4 5 6 7 8 9 10 data(ctd) plot(ctd, which="N2") rho <- swRho(ctd) z <- swZ(ctd) zz <- seq(min(z), max(z), 0.1) N2 <- -9.8 / mean(rho) * runlm(z, rho, zz, deriv=1) lines(N2, -zz, col='red') legend("bottomright", lwd=2, bg="white", col=c("black", "red"), legend=c("swN2()", "using runlm()"))
The fit in Case 1 is almost spookily good.
The N^2 results suggest including this as a method for
swN2(), perhaps the default method, but that’s for another day.
- Source code: 2014-04-11-runlm.R