Description Usage Arguments Details Value Examples
View source: R/mcmc_agric_model.R
1 2 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 | mcmc_agric_model(
data_arc,
data_mod,
impute_spec = c("full", "cut", "smi")[1],
power_w_PO = 1,
power_w_HM = 1,
prior_spec_PO = c("flat", "proper")[2],
prior_spec_HM = c("flat", "proper")[2],
PO_site_rnd_eff = TRUE,
HM_site_rnd_eff = TRUE,
n_iter = 10000,
n_iter_sub = 100,
n_warmup = 1000,
n_thin = 2,
theta_ini = NULL,
theta_min_max = NULL,
theta_prop_int = NULL,
theta_prop_kernel = c("norm", "unif")[1],
n_epoch_adapt = 0,
n_iter_adapt = 2000,
ManureLevel_arc_imp_ini = NULL,
Rainfall_arc_imp_ini = NULL,
imp_playpen = FALSE,
gibbs_hm = TRUE,
PO_expand = FALSE,
POmixda = TRUE,
lambda_prop_int = 0.2,
lambda_mix_PO_prior = c(1, 1),
keep_imp = FALSE,
keep_ll = FALSE,
elpd = FALSE,
out_file_rda = NULL,
log_file = NULL,
devel = FALSE
)
|
data_arc |
Data frame. Archaeological data |
data_mod |
Data frame. Modern data |
impute_spec |
Character. Specification of imputed values of ManureLevel: "full", "cut", "smi" |
power_w_PO |
Numeric. Raise the likelihood in the PO module to a power when performing M-H steps. |
power_w_HM |
Numeric. Raise the likelihood in the HM module to a power when performing M-H steps. |
prior_spec_PO |
specification of prior distributions for the parameters in the PO model: "flat" or "proper" |
prior_spec_HM |
specification of prior distributions for the parameters in the HM model: "flat" or "proper" |
PO_site_rnd_eff |
Boolean. Shall the PO module use random effects by Site |
HM_site_rnd_eff |
Boolean. Shall the HM module use random effects by Site |
n_iter |
Integer. Number of iterations in the main MCMC chain. |
n_iter_sub |
Integer. Number of updates in the subchain for parameters in the PO module. |
n_warmup |
Integer. Number of updates discarded when "warming-up" the MCMC |
n_thin |
Integer. One of every n_thin updates will be kept (thinning the chain) |
theta_ini |
Numeric vector. Initial values for the parameters |
theta_min_max |
matrix with two columns, minimum and maximum values for each parameter |
theta_prop_int |
Used to control the width of the proposal distribution for parameters in PO and HM modules |
theta_prop_kernel |
Shape of the proposal distribution: "uniform" or "normal" |
n_epoch_adapt |
Integer. Number of epochs that adaptation runs of the MCMC, to addapt the proposal distribution, is performed before the real chain. |
n_iter_adapt |
Integer. Number of iterations in the adaptation runs of the MCMC. |
ManureLevel_arc_imp_ini |
Initial values for imputing the missing values of Rainfall |
Rainfall_arc_imp_ini |
Initial values for imputing the missing values of Rainfall |
imp_playpen |
(Experimental). If True, The missing manure levels are imputed in to avoid quasi-complete separation in the PO module. |
gibbs_hm |
Boolean. Shall we use Gibss to update parameters in the HM module. If FALSE, M_H is used. |
PO_expand |
Indicates if the Proportional Odds models should be expanded by considering an additional mixture component |
POmixda |
Indicates if the inference for the mixture model uses data augmentation |
lambda_prop_int |
Double. Width of the proposal distribution for the mixing weight, when PO_expand=TRUE |
lambda_mix_PO_prior |
Numeric vector. indicates the two parameters of the beta prior for the mixture weight |
keep_imp |
Indicates if the imputed values for missing data should be returned |
keep_ll |
Indicates if the individual log-likelihoods should be returned |
elpd |
Indicates if the ELPD should be computed and returned |
out_file_rda |
Indicates a file (.rda) where the output should be saved |
log_file |
Indicates a file (.txt) where the log of the process should be saved |
devel |
Development mode |
The hierarchical model consists of two parts: The Proportional Odds (PO1) component, and the Gaussian Linear model (HM1). HM1: normd15N ∼ 1 + Rainfall + ManureLevel + (1|Site) weights = nlme::varIdent (form=~1|Category) PO1: ManureLevel ∼ 1 + Size + (1|Site) link = "logistic"
MCMC details: Coefficients in the HM1 modulel can be updated using Gibbs sampling (gibbs_hm=TRUE) for the joint conditional posterior, M-H otherwise Parameters in the PO module are updated using M-H, updating one by one. ManureLevel missing values is updated using M-H one value at a time. Rainfall missing values are updated all together using M-H.
1 2 3 4 5 6 7 8 9 | This function allows to perform adaptations for the proposal of the PO1 parameters. Change n_epoch_adapt and n_iter_adapt.
In this implementation, we allow to perform several types of inference.
1) Conventional Bayes (default): impute_spec="full", set power_w_HM=1, power_w_PO=1
2) Powered likelihood: impute_spec="full", gibbs_hm="FALSE", set "power_w_HM" and "power_w_PO" to control the influence of each module in the update of parameters and missing data.
3) Cut model: impute_spec="cut" (deprecate set power_w_HM and power_w_PO). Bayesian multiple imputation for ManureLevel.
4) smi imputation: impute_spec="smi", set "power_w_HM" and "power_w_PO" to control the influence of each module in the imputation of ManureLevel.
Parameters are initialized in the MLE using a single imputation of missing values.
|
A list with three main elements, described below, and some details about the run.
theta_mcmc
Matrix with the chains of the parameters in the model.
ManureLevel_imp_mcmc
if keep_imp=TRUE, matrix with the chains of the imputed values of the missing ManureLevel.
Rainfall_imp_mcmc
if keep_imp=TRUE, matrix with the chains of the imputed values of the missing Rainfall.
1 2 3 4 5 6 | ## Not run:
##### Cut model for NMeso data #####
## End(Not run)
|
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.