Given model and data, this function a Hamiltonian MCMC sampler and stores the samples in an mcmc.list called samples.
Posterior predictive checks developed by Klauer (2010), deviance information criterion (DIC; Spiegelhalter et al., 2002),
99% and 95% highest density intervals (HDI) together with the median will be provided for the main parameters in a list
called diags. Optionally, the indices widely applicable information criterion (WAIC; Watanabe, 2010; Vehtari et al., 2017) and
leave-one-out cross-validation (LOO; Vehtari et al., 2017) can be saved. Additionally the log-likelihood (LogLik) can also be stored.
Some specifications of the function call are also saved in specs.
fit_drtmpt(
model,
data,
n.chains = 4,
n.iter = 1000,
n.phase1 = 1000,
n.phase2 = 2000,
n.thin = 1,
Rhat_max = 1.1,
Irep = 1000,
prior_params = NULL,
flags = NULL,
control = NULL
)A list of the class drtmpt_fit containing
samples: the posterior samples as an mcmc.list object,
diags: some diagnostics like deviance information criterion, posterior predictive checks for the frequencies and latencies,
potential scale reduction factors, and also the 99% and 95% HDIs and medians for the group-level parameters,
specs: some model specifications like the model, arguments of the model call, and information about the data transformation,
indices (optional): if enabled, WAIC and LOO,
LogLik (optional): if enabled, the log-likelihood matrix used for WAIC and LOO.
summary includes posterior mean and median of the main parameters.
A list of the class drtmpt_model.
Optimally, a list of class drtmpt_data. Also possible is a data.frame or a
path to the text file. Both, data.frame and the text file must contain the column names "subj",
"group", "tree", "cat", and "rt" preferably but not necessarily in this order. The values of the latter must
be in milliseconds. It is always advised to use to_drtmpt_data first, which gives back a
drtmpt_data list with information about the changes in the data, that were needed.
Number of chains to use. Default is 4. Must be larger than 1 and smaller or equal to 16.
Number of samples per chain. Default is 1000.
Number of samples for phase 1 (adaptation phase). Default is 1000.
Number of samples for phase 2. Default is 2000.
Thinning factor. Default is 1.
Maximal Potential scale reduction factor: A lower threshold that needs to be reached before the actual sampling starts. Default is 1.05
Every Irep samples an interim state with the current maximal potential scale reduction
factor is shown. Default is 1000. The following statements must hold true for Irep:
n.phase1 is a multiple of Irep
n.phase2 is a multiple of Irep
n.phase1 is smaller than or equal to n.phase2,
Irep is a multiple of n.thin and
n.iter is a multiple of Irep / n.thin.
Named list with prior parameters. All parameters have default values, that lead to uninformative priors. Vectors are not allowed. Allowed parameters are:
prec_epsilon: prior precision for population means of process-related parameters.
Default is 1.0.
delta_df: degrees of freedom of t-distribution for motor times. Default is 10.
delta_mu: mean of t-distribution for motor times. Default is 0.5.
delta_scale: scale of t-distribution for motor times. Default is 1.0.
SIGMA_Corr_eta: shape parameter for LKJ distribution for process-related parameters. Default is 4.0.
SIGMA_SD_rho: scale parameter of half-Cauchy distribution for process-related parameters. Default is 2.5.
GAMMA_Corr_eta: shape parameter for LKJ distribution for motor-related parameters. Default is 4.0.
GAMMA_SD_rho: scale parameter of half-Cauchy distribution for motor-related parameters. Default is 0.5.
Omega2_alpha: shape parameter of gamma distribution for residual variance. Default is 0.0025.
Omega2_beta: rate parameter of gamma distribution for residual variance. Default is 0.5.
Either NULL or a list of
old_label If set to TRUE the old labels of "subj" and "group" of the data will be used in the elements of the output list.
Default is FALSE.
indices Model selection indices. If set to TRUE the log-likelihood for each iteration and trial will be
stored temporarily and with that the DIC, as well as the WAIC and LOOIC will be calculated via the loo package. If you want to have the
log-likelihood matrix stored in the output of this function, you can set loglik to TRUE. Default for
indices is FALSE.
loglik If set to TRUE and indices = TRUE the log-likelihood matrix for each iteration and trial will
be saved in the output as a matrix. Default is FALSE.
random_init If set to TRUE the initial values are randomly drawn. If FLASE maximum likelihood is used
for initial values.
Either NULL or a list of
maxthreads for the ML estimation of the initial values and the calculation of the DIC values one can use more than
n.chains threads for parallelization. Default is 4 like n.chians. maxthreads must be larger or equal to
n.chains.
maxtreedepth1_3 maxtree-depth of the no-U-turn algorithm in Phases 1 to 3
maxtreedepth4 maxtree-depth of the no-U-turn algorithm in Phases 4
Raphael Hartmann
Klauer, K. C. (2010). Hierarchical multinomial processing tree models: A latent-trait approach. Psychometrika, 75(1), 70-98.
Spiegelhalter, D. J., Best, N. G., Carlin, B. P., & Van Der Linde, A. (2002). Bayesian measures of model complexity and fit. Journal of the royal statistical society: Series b (statistical methodology), 64(4), 583-639.
Vehtari, A., Gelman, A., & Gabry, J. (2017). Practical Bayesian model evaluation using leave-one-out cross-validation and WAIC. Statistics and Computing, 27(5), 1413-1432.
Watanabe, S. (2010). Asymptotic equivalence of Bayes cross validation and widely applicable information criterion in singular learning theory. Journal of Machine Learning Research, 11(Dec), 3571-3594.
####################################################################################
# Detect-Guess variant of the Two-High Threshold model.
# The encoding and motor execution times are assumed to be equal for each response.
####################################################################################
mdl_2HTM <- "
# targets
do+(1-do)*g
(1-do)*(1-g)
# lures
(1-dn)*g
dn+(1-dn)*(1-g)
# do: detect old; dn: detect new; g: guess
"
model <- to_drtmpt_model(mdl_file = mdl_2HTM)
data_file <- system.file("extdata/data.txt", package="rtmpt")
data <- read.table(file = data_file, header = TRUE)
data_list <- to_drtmpt_data(raw_data = data, model = model)
# \donttest{
# This might take some time
drtmpt_out <- fit_drtmpt(model = model, data = data_list, Rhat_max = 1.1)
drtmpt_out
# }
Run the code above in your browser using DataLab