Description Usage Arguments Details Value Note Author(s) References
View source: R/dream_parallel.R
Implementation of the DREAM algorithm, including variations (such as DREAM(zs) and DEAM(ABC)) depending on certain argument settings.
1 2 3 4 5 6 7 8 9 | dream_parallel(fun, ..., lik = NULL, par.info = list(initial = NULL, min =
NULL, max = NULL, mu = NULL, cov = NULL, val_ini = NULL, bound = NULL, names =
NULL, prior = "flat"), nc, t, d, burnin = 0, adapt = 0.1,
updateInterval = 10, delta = 3, c_val = 0.1, c_star = 1e-12,
nCR = 3, p_g = 0.2, beta0 = 1, thin = 1, outlier_check = TRUE,
obs = NULL, abc_rho = NULL, abc_e = NULL, glue_shape = NULL,
lik_fun = NULL, past_sample = FALSE, m0 = NULL, archive_update = NULL,
psnooker = 0, mt = 1, ncores = 1, checkConvergence = FALSE,
verbose = TRUE)
|
fun |
|
... |
Additional arguments for |
lik |
|
par.info |
A |
nc |
|
t |
|
d |
|
burnin |
|
adapt |
|
updateInterval |
|
delta |
|
c_val |
|
c_star |
|
nCR |
|
p_g |
|
beta0 |
|
thin |
|
outlier_check |
|
obs |
|
abc_rho |
|
abc_e |
|
glue_shape |
|
lik_fun |
|
past_sample |
|
m0 |
|
archive_update |
|
psnooker |
|
mt |
|
ncores |
|
checkConvergence |
|
verbose |
|
initial |
|
min |
|
max |
|
mu |
|
cov |
|
val_ini |
|
bound |
|
names |
|
prior |
|
To understand the notation (e.g. what is lambda, nCR etc.), have a look at Sect. 3.3 of the reference paper (see below).
Likelihood options (argument lik
):
1: fun
returns the likelihood of parameter realisation x given some observations.
2: fun
returns the log-likelihood.
21: ABC diagnostic model evaluation using a continuous fitness kernel, a variation of so-called noisy-ABC.
22: ABC diagnostic model evaluation using a Boxcar likelihood. fun
has to return m diagnostic values
to be compared with observations. Requires arguments obs
, abc_rho
, and abc_e
. Modification
in computation of Metropolis probability is used (Eq. 13 of Sadegh and Vrugt, 2014). Prior pdf set to zero!
31: GLUE with informal likelihood function based on the NSE: glue_shape * log(NSE)
. fun
needs to
return a time series of model simulations and obs
should contain a time series of corresponding observations.
glue_shape
affects the sampling: small values result in high parameter uncertainty, large values produce
narrow posterior pdfs of parameters. Should be in the range of 1-100 (experiment!).
99: A user-defined function, see argument lik_fun
.
list
with named elements:
chain: a (1-burnin)*t/thin-by-d+3-by-nc array of parameter realisations and the log-pdfs of the prior ('lp'), likelihood ('ll'), and posterior ('lpost') distribution for each iteration and Markov chain;
fx: a (1-burnin)*t/thin-by-nc-by-[length of fun
's output] array of raw output of fun
,
corresponds with parameter realisations in chain;
runtime: time of function execution;
outlier: a list with adapt*t vectors of outlier indices in nc (value of 0 means no outliers);
AR: a [floor(t/updateInterval)+1
]-by-2 matrix giving the total number of proposal evaluations an the
associated acceptance rate averaged over the last updateInterval * nc
evaluations.
CR: a [floor(t/updateInterval)+1
]-by-[nCR+1] matrix giving the total number of proposal evaluations
(first column) an the current selection probability for each of the nCR
crossover values.
IF checkConvergence == TRUE:
R_stat: a (1-burnin)*t/thin-50+1-by-d matrix giving the Gelman-Rubin convergence diagnostic (note that at least 50 observations are used to compute R_stat).
If you want to use a non-implemented likelihood function with nuisance variables to be calibrated along
with the actual model parameters, it is suggested to implement the likelihood calculation directly into fun
.
Tobias Pilz tpilz@uni-potsdam.de
Code based on:
Vrugt, J. A.: "Markov chain Monte Carlo simulation using the DREAM software package: Theory, concepts, and MATLAB implementation." Environmental Modelling & Software, 2016, 75, 273 – 316, http://dx.doi.org/10.1016/j.envsoft.2015.08.013.
For ABC method see:
Sadegh, M. and J. A. Vrugt: "Approximate Bayesian computation using Markov Chain Monte Carlo simulation: DREAM_ABC". Water Resources Research, 2014, 50, 6767 – 6787, http://dx.doi.org/10.1002/2014WR015386.
For MT-DREAM(ZS) see:
Laloy, E. and J. A. Vrugt: "High-dimensional posterior exploration of hydrologic models using multiple-try DREAM(ZS) and high-performance computing". Water Resources Research, 2012, 48, W01526, http://dx.doi.org/10.1029/2011WR010608.
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.