You ask whether there would be a way to adapt the deriv and
deriv3 functions to deal with formulas that have an indexed term,
effectively allowing differentiation with respect to a vector. As Simon
Blomberg famously said, This is R. There is no if. Only how.(*)
The implementation,
I have found a way to approach the question that I posed, that is
having deriv and deriv3 deal with an indexed term in the formula.
Unless, there is a more direct approach (and my minimal investigation
of the code involved suggested that it would not be simple to implement,
for me at least), this
A slight glitch got-in during the copy and past.
The part below that says,
ms.curv(dd.nls.d4)
Parameter effects: c^theta x sqrt(F) = 0.0123
Intrinsic: c^iota x sqrt(F) = 0.0068
should read,
rms.curv(dd.nls.d4)
Parameter effects: c^theta x sqrt(F) = 0.0123
Intrinsic: c^iota x
This works for me in terms of giving results without error messages
except for the confint(dd.plin) which I assume you don't really need anyways.
gg - model.matrix(~ Gun/GL - Gun, dd)
dd.plin - nls(Lum ~ gg^gamm, dd, start = list(gamm = 2.4),
+alg = plinear
+
I mixed up examples. Here it is again. As with the last one
confint(dd.plin) gives an error (which I assume is a problem with
confint that needs to be fixed) but other than that it works without
issuing errors and I assume you don't need the confint(dd.plin)
in any case since dd.plin is just
I don't know why I get a different result than you do, though I'll
play around with it some and check the code to see if I can figure
out what is going on, on my end.
I will add that by transforming GL so that it spans the interval [0, 1]
instead
of [0, 255], the correlations among the parameters
Hi,
I'm fitting a standard nonlinear model to the luminances measured
from the red, green and blue guns of a TV display, using nls.
The call is:
dd.nls - nls(Lum ~ Blev + beta[Gun] * GL^gamm,
data = dd, start = st)
where st was initally estimated using optim()
st
$Blev
Please provide reproducible code which shows the error.
dd - structure(list(Lum = c(0.15, 0.07, 0.1, 0.19, 0.4, 0.73, 1.2,
+ 1.85, 2.91, 3.74, 5.08, 6.43, 8.06, 9.84, 12, 14.2, 16.6, 0.1,
+ 0.1, 0.17, 0.46, 1.08, 2.22, 3.74, 5.79, 8.36, 11.6, 15.4, 19.9,
+ 24.6, 30.4, 36.1, 43, 49.9, 0.06, 0.06,
Thank you for your rapid response.
This is reproducible on my system. Here it is again, with, I hope,
sufficient detail to properly document what does not work and what
does on my system,
But my original question, properly motivated or not, concerns whether
there is a way to use or adapt
Sorry, a slight glitch got in because I subsetted the original data frame.
The last part, that works should read:
dd.nls.d2 - nls(Lum ~ dd.deriv2(Blev, beta, gamm, GL), data = dd,
start = list(Blev = B[2], beta = B[3:5], gamm = B[1]))
#But not here
confint(dd.nls.d2)
10 matches
Mail list logo
