Closed kylebaron closed 1 year ago
From NM-Help; these are the valid values for METHOD
in $ESTIM
; I believe that METHOD=CHAIN
is the only one that doesn't do some estimation.
METHOD=kind
Values for kind are:
0 or ZERO
Always set etas to 0 during the computation of the objective
function. Also called the "first order (FO) method." This
is the default.
1 or CONDITIONAL
Use conditional estimates for the etas during the computa-
tion of the objective function. METHOD=1 (without the
LAPLACIAN option) is also called the "first order condi-
tional estimation (FOCE) method." The conditional estimates
of the etas are referred to as Conditional Parametric Etas
(CPE).
METH=COND NOLAPLACIAN is referred to as the FOCE method.
METH=COND LAPLACE is referred to as the Laplace method.
METH=COND NOLAPLACE CENTERING is referred to as the Centering FOCE method.
METH=COND LAPLACE CENTERING is referred to as the Centering Laplace method.
HYBRID
Use conditional estimates for the etas during the computa-
tion of the objective function, with the exception of those
etas listed in the ZERO option. Cannot be used with LAPLA-
CIAN or CENTERING.
The following methods are new to NONMEM 7. When any of these
methods are used, the data are inferred to be population, and
METHOD=1 is supplied if it is not already present. The first
four methods are referred to as EM (Expectation-Maximization)
Methods.
ITS Use the iterative two-stage method. This method evaluates
the conditional mode and first order approximation of the
conditional variance of parameters of individuals by maxi-
mizing the posterior density. This integration step is the
same as used in the FOCE method. Population parameters are
updated from individuals' conditional mode parameters and
their approximate variances by single iteration maximization
steps.
IMP Use the Monte-Carlo Importance Sampling Expectation Maxi-
mization method. This method evaluates the conditional mean
and variance of parameters of individuals by Monte Carlo
sampling. It uses the posterior density, which incorporates
the likelihood of parameters relative to population means
and variances with the individual's observed data. The nor-
mal density near the mean or mode of the posterior is used
as a proposal density, then weighted according to the poste-
rior density as a correction.
IMPMAP
Use the Importance Sampling method assisted by Mode a Poste-
riori (MAP) estimation. At each iteration, conditional
modes and conditional first order variances are evaluated as
in the ITS or FOCE method. These are then used as parame-
ters to the multivariate normal proposal density for the
Monte-Carlo importance sampling step.
SAEM Use the Stochastic Approximation Expectation Maximization
method. As in importance sampling, random samples are gen-
erated from normal proposal densities. However, instead of
always being centered at the mean or mode of the posterior
density, the proposal density is centered at the previous
sample position.
BAYES
Use the Markov Chain Monte Carlo (MCMC) Bayesian Analysis
method. The goal of the MCMC Bayesian analysis is to obtain
a large sample set of probable population parameters. Vari-
ous summary statistics of the population parameters may then
be obtained such as means and confidence ranges.
DIRECT
Requests Monte Carlo Direct Sampling. Creates completely
independent samples (unlike MCMC), and there is no chance of
causing bias if the sampling density is not similar enough
to the conditional density (unlike IMP). However, it is
very inefficient, requiring ISAMPLE settings of 10000 to
300000 to properly estimate the problem.
NUTS (NM74)
Requests No U-Turn Sampling (NUTS) Markov Chain Monte Carlo
(MCMC) Bayesian Analysis Method. Options unique to this
method are listed alphabetically under NUTS_.... Other
options of interest with their defaults are as follows:
MASSRESET=-1
MADAPT=-1
KAPPA=1
TTDF=0
OLKJDF=0
OVARF=1
SLKJDF=0
SVARF=1
CHAIN
Allows the user to create a series of random initial values
of THETAs and OMEGA's, or for reading in initial population
parameters from a file of rectangular (rows/column) format.
Applies only to the Estimation Step.
Summary
Ref: https://github.com/metrumresearchgroup/bbi/pull/272/files#r965087052
When we run Bayesian estimation, we use two
$ESTIMATION
recordsThe
METHOD=CHAIN
record does not produce any parameter estimates.So you will find two estimation methods in the
.lst
file (when looking for#METH
), but there will only be one set of estimates.This setup currently produces an error: