Closed BertvanderVeen closed 3 years ago
Likely duplicate of #189 (does it solve the problem to set openmp(1)
?).
matern
calls the special functions lgamma
and BesselK
which both might throw warnings for extreme (large or small) parameters. Until #189 has been solved a possible workaround is to add a validpar
implementation to your obj - see ?MakeADFun
.
That makes sense, exp(lg_phi)
tends to infinity for my example. Setting openmp(1)
does not solve the problem (in addition, it happens on vanilla R too). Thanks for the workaround, but do you happen to have a recommendation of a valid parameter range too?
but do you happen to have a recommendation of a valid parameter range too?
No, it's problem specific
Description:
While attempting to fit a spatial model in TMB, with matern covariance, R crashes when maximizing. When switching from matern to exp(-phi*H(i,j)) things work fine.
Reproducible Steps:
Current Output:
Expected Output:
-
TMB Version:
1.7.16
R Version:
4.0.2,3.6.2, with microsoft R open with intel MKL
Operating System:
Windows 10, Ubuntu 18.04