The problem lies in dgamma, the function which is stated to be implicated.
See the following NEWS item:
o [dpqr]gamma now returns NaN for an invalid 'shape' parameter
(rather than throw an error), for consistency with other
distribution functions.
That setting lower/upper w
Thanks to both Peters for involvment. I will add argument for method to
be used in optim.
--
Lep pozdrav / With regards,
Gregor Gorjanc
--
University of Ljubljana PhD student
Biotechnical Faculty
Zootechnical Department
I think the problem may lie with fitdistr().
Specifically, replacing the code in fitdistr.R (VR_7.2-20)
(line 137 to end) with the code in VR_7.2-8 (line 92 to end)
seems to handle
fitdistr(otm, "gamma")
just fine. But I haven't done much testing.
Peter Ehlers
Peter Dalgaard wrote:
> P Ehler
P Ehlers <[EMAIL PROTECTED]> writes:
> Gregor,
>
> fitdistr(otm, "gamma", method="L-BFGS-B")
>
> works for me (on WinXP). Or you could specify "lower = 0".
The really odd thing is that it even works with
> fitdistr(otm, "gamma",lower=-Inf)
shape rate
1.03081094 0.18924370
Gregor,
fitdistr(otm, "gamma", method="L-BFGS-B")
works for me (on WinXP). Or you could specify "lower = 0".
I no longer have 2.1.0 running, so I don't know why this
wasn't needed in 2.1.0.
"R version 2.2.0, 2005-10-24"
MASS version: 7.2-20
-peter
Gorjanc Gregor wrote:
> Dear R developers,
Dear R developers,
I have encountered strange behaviour of fitdistr for gamma in recent R
build i.e. 2.2.0. I have attached the code for data at the end of this mail
so you can reproduce the problem. In short, I am able to run fitdistr under
2.1.0 without problems, while I get the following error
