Description Usage Arguments Details Value Examples
Draw MCMC samples from a TMB model object using Stan
1 2 |
obj |
TMB model object. |
... |
Passed to |
lower |
Vector of lower parameter bounds. |
upper |
Vector of upper parameter bounds. |
laplace |
Apply the Laplace approximation to |
silent |
Be silent during samling ? |
debug |
Should not be used. |
tmbstan
works for models with or without random effects.
By default a full Bayesian analysis is carried out, i.e. both
parameters and random effects are sampled using MCMC. Models with
random effects will thus have the Laplace approximation disabled. It
is possible to mix the Laplace approximation with MCMC by setting
laplace=TRUE
.
All methods provided by the rstan
package can be applied to a
fitted object. Get a complete list using
methods(class="stanfit")
.
Lower and upper bounds can be set using lower
and upper
.
The bounds can be specified in one of two ways. Either in short
format, i.e. have the same length as obj$par
. Remaining
parameters (the random effects) are set as unbounded in this case.
Otherwise the bounds must be in long format, i.e. have the same
length as the full parameter vector obj$env$par
including the
random effects.
In both cases -Inf
and Inf
are valid components of
lower
and upper
respectively.
Note that initial values must be within the specified bounds.
The function arguments ...
are passed to rstan
's
fitting function, see ?rstan::sampling
.
A few notable arguments are:
chains
The number of chains.
iter
The number of iterations.
init
Initial values for the sampler.
Behaves like rstan
with some additions:
Default is "random"
- see ?stan
.
Special values 0
and "0"
are allowed - see ?stan
.
Additional special characters "par"
and "last.par.best"
are allowed
and will be looked up in the TMB model object.
The value "par"
signifies to start from the defaults of the model object.
If an optimization has been carried out, the intial value "last.par.best"
will start from the MLE.
We also allow to pass a single numeric vector, or a list of numeric vectors. List length must match the number of chains. Vector lengths must match the number of sampled parameters. Names are currently ignored.
Parameters
that do not follow the previous scheme (e.g. characters) are passed on
to rstan
unchanged. If in doubt, use rstan::get_inits
to inspect the
applied initial values.
seed
Random seed.
Object of class stanfit
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 |
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.