Nothing
R/call_rtmpt.R
In rtmpt: Fitting (Exponential/Diffusion) RT-MPT Models
Defines functions
fit_drtmpt
fit_ertmpt
Documented in
fit_drtmpt
fit_ertmpt
#' Fit Exponential-RT-MPT Models
#'
#' Given model and data, this function calls an altered version of the C++ program by Klauer and Kellen (2018) to sample from
#' the posterior distribution via a Metropolis-Gibbs sampler and storing it in an mcmc.list called \code{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 \code{diags}. Optionally, the \code{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 (\code{LogLik}) can also be stored.
#' Some specifications of the function call are also saved in \code{specs}.
#'
#' @param model A list of the class \code{ertmpt_model}.
#' @param data Optimally, a list of class \code{ertmpt_data}. Also possible is a \code{data.frame} or a
#' path to the text file. Both, \code{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 \code{\link{to_ertmpt_data}} first, which gives back an \code{ertmpt_data} list
#' with informations about the changes in the data, that were needed.
#' @param n.chains Number of chains to use. Default is 4. Must be larger than 1 and smaller or equal to 16.
#' @param n.iter Number of samples per chain. Default is 5000.
#' @param n.burnin Number of warm-up samples. Default is 200.
#' @param n.thin Thinning factor. Default is 1.
#' @param Rhat_max Maximal Potential scale reduction factor: A lower threshold that needs to be reached before the actual sampling starts. Default is 1.05
#' @param Irep Every \code{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 \code{Irep}:
#' \itemize{
#' \item \code{n.burnin} is smaller than or equal to \code{Irep},
#' \item \code{Irep} is a multiple of \code{n.thin} and
#' \item \code{n.iter} is a multiple of \code{Irep / n.thin}.
#' }
#' @param prior_params Named list with prior parameters. All parameters have default values, that lead to uninformative priors. Vectors are not allowed.
#' Allowed parameters are:
#' \itemize{
#' \item \code{mean_of_exp_mu_beta}: This is the a priori expected exponential rate (\code{E(exp(beta)) = E(lambda)}) and
#' \code{1/mean_of_exp_mu_beta} is the a priori expected process time (\code{1/E(exp(beta)) = E(tau)}). The default
#' mean is set to \code{10}, such that the expected a priori process time is \code{0.1} seconds.
#' \item \code{var_of_exp_mu_beta}: The a priori group-specific variance of the exponential rates. Since
#' \code{exp(mu_beta)} is Gamma distributed, the rate of the distribution is just mean divided by variance and
#' the shape is the mean times the rate. The default is set to \code{100}.
#' \item \code{mean_of_mu_gamma}: This is the a priori expected \emph{mean parameter} of the encoding and response execution times,
#' which follow a normal distribution truncated from below at zero, so \code{E(mu_gamma) < E(gamma)}. The default is \code{0}.
#' \item \code{var_of_mu_gamma}: The a priori group-specific variance of the \emph{mean parameter}. Its default is \code{10}.
#' \item \code{mean_of_omega_sqr}: This is the a priori expected residual variance (\code{E(omega^2)}). Its distribution
#' differs from the one used in the paper. Here it is a Gamma distribution instead of an improper one. The default
#' is \code{0.005}.
#' \item \code{var_of_omega_sqr}: The a priori variance of the residual variance (\code{Var(omega^2)}). The default is
#' \code{0.01}. The default of the mean and variance is equivalent to a shape and rate of \code{0.0025} and
#' \code{0.5}, respectivly.
#' \item \code{df_of_sigma_sqr}: A priori degrees of freedom for the individual variance of the response executions. The
#' individual variance has a scaled inverse chi-squared prior with \code{df_of_sigma_sqr} degrees of freedom and
#' \code{omega^2} as scale. \code{2} is the default and it should be an integer.
#' \item \code{sf_of_scale_matrix_SIGMA}: The original scaling matrix (S) of the (scaled) inverse Wishart distribution for the process
#' related parameters is an identity matrix \code{S=I}. \code{sf_of_scale_matrix_SIGMA} is a scaling factor, that scales this
#' matrix (\code{S=sf_of_scale_matrix_SIGMA*I}). Its default is \code{1}.
#' \item \code{sf_of_scale_matrix_GAMMA}: The original scaling matrix (S) of the (scaled) inverse Wishart distribution for the encoding and
#' motor execution parameters is an identity matrix \code{S=I}. \code{sf_of_scale_matrix_GAMMA} is a scaling factor, that scales
#' this matrix (\code{S=sf_of_scale_matrix_GAMMA*I}). Its default is \code{1}.
#' \item \code{prec_epsilon}: This is epsilon in the paper. It is the precision of mu_alpha and all xi (scaling parameter
#' in the scaled inverse Wishart distribution). Its default is also \code{1}.
#' \item \code{add_df_to_invWish}: If \code{P} is the number of parameters or rather the size of the scale matrix used in the (scaled)
#' inverse Wishart distribution then \code{add_df_to_invWish} is the number of degrees of freedom that can be added to it. So
#' \code{DF = P + add_df_to_invWish}. The default for \code{add_df_to_invWish} is \code{1}, such that the correlations are uniformly
#' distributed within \code{[-1, 1]}.
#' }
#' @param indices Model selection indices. If set to \code{TRUE} the log-likelihood for each iteration and trial will be stored temporarily
#' and with that the WAIC and LOO will be calculated via the \code{loo} package. If you want to have this log-likelihood matrix stored in the
#' output of this function, you can set \code{save_log_lik} to \code{TRUE}. The default for \code{indices} is \code{FALSE}.
#' @param save_log_lik If set to \code{TRUE} the log-likelihood matrix for each iteration and trial will
#' be saved in the output as a matrix. Its default is \code{FALSE}.
#' @param old_label If set to \code{TRUE} the old labels of "subj" and "group" of the data will be used in the elements of the output list. Default is \code{FALSE}.
#' @return A list of the class \code{ertmpt_fit} containing
#' \itemize{
#' \item \code{samples}: the posterior samples as an \code{mcmc.list} object,
#' \item \code{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,
#' \item \code{specs}: some model specifications like the model, arguments of the model call, and information about the data transformation,
#' \item \code{indices} (optional): if enabled, WAIC and LOO,
#' \item \code{LogLik} (optional): if enabled, the log-likelihood matrix used for WAIC and LOO.
#' \item \code{summary} includes posterior mean and median of the main parameters.
#' }
#' @references
#' Hartmann, R., Johannsen, L., & Klauer, K. C. (2020). rtmpt: An R package for fitting response-time extended multinomial processing tree models.
#' \emph{Behavior Research Methods, 52}(3), 1313–1338.
#'
#' Hartmann, R., & Klauer, K. C. (2020). Extending RT-MPTs to enable equal process times. \emph{Journal of Mathematical Psychology, 96}, 102340.
#'
#' Klauer, K. C. (2010). Hierarchical multinomial processing tree models: A latent-trait approach. \emph{Psychometrika, 75(1)}, 70-98.
#'
#' Klauer, K. C., & Kellen, D. (2018). RT-MPTs: Process models for response-time distributions based on multinomial processing trees with
#' applications to recognition memory. \emph{Journal of Mathematical Psychology, 82}, 111-130.
#'
#' Spiegelhalter, D. J., Best, N. G., Carlin, B. P., & Van Der Linde, A. (2002). Bayesian measures of model complexity and fit.
#' \emph{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.
#' \emph{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.
#' \emph{Journal of Machine Learning Research, 11(Dec)}, 3571-3594.
#' @examples
#' ####################################################################################
#' # 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_ertmpt_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_ertmpt_data(raw_data = data, model = model)
#' \donttest{
#' # This might take some time
#' ertmpt_out <- fit_ertmpt(model = model, data = data_list, Rhat_max = 1.1)
#' ertmpt_out
#' }
#' # Type ?SimData for another working example.
#' @author Raphael Hartmann
#' @useDynLib "rtmpt", .registration=TRUE
#' @export
#' @importFrom coda gelman.diag
#' @importFrom methods as callGeneric new is
#' @importFrom utils read.table write.table
fit_ertmpt <- function(model,
data,
n.chains = 4,
n.iter = 5000,
n.burnin = 200,
n.thin = 1,
Rhat_max = 1.05,
Irep = 1000,
prior_params = NULL,
indices = FALSE,
save_log_lik = FALSE,
old_label = FALSE) {
# internal argument name changes
Nchains <- n.chains
Nsamples <- n.iter
Nwarmup <- n.burnin
thin <- n.thin
# some controls
model_elmnts <- c("lines", "params", "responses")
if (!is.list(model)) stop("\"model\" must be a list.")
if (!all(model_elmnts %in% names(model))) stop("\"model\" must contain \"", model_elmnts[which(!(model_elmnts %in% names(model)))[1]], "\".")
if (Nchains < 2 || Nchains > 16) stop("\"n.chains\" must be larger than 1 and lower or equal to 16.")
if (Irep < 1) stop("\"Irep\" must be larger than or equal to one.")
if (thin < 1) stop("\"n.thin\" must be larger than or equal to one.")
if (Irep %% thin != 0) stop("\"Irep\" must be a multiple of \"n.thin\".")
if (Nsamples %% (Irep/thin) != 0) stop("\"n.iter\" must be a multiple of \"Irep\" / \"n.thin\".")
if (Nsamples < Irep/thin) stop("\"n.iter\" must be greater or equal to \"Irep\" / \"n.thin\" = ", Irep*thin, ".")
if (Nwarmup > Irep) stop("\"n.burnin\" must be smaller than or equal to \"Irep\".")
if (Rhat_max < 1) stop("\"Rhat_max\" must be larger than or equal to one.")
if (!is.logical(indices)) stop("\"indices\" must either be TRUE or FALSE.")
# if (!is.logical(bridge)) stop("\"bridge\" must either be TRUE or FALSE.")
bridge = FALSE
if (!is.logical(save_log_lik)) stop("\"save_log_lik\" must either be TRUE or FALSE.")
if (!is.logical(old_label)) stop("\"old_label\" must either be TRUE or FALSE.")
if (!is.null(prior_params) && !is.list(prior_params)) stop("\"prior_params\" must be a list.")
# prepare data
keep_data_path <- FALSE
if (is.data.frame(data)) {
temp_data <- to_ertmpt_data(data, model)
data_frame <- temp_data$data
if("transformation" %in% names(temp_data)) {
transformation <- temp_data$transformation
} else transformation <- list()
} else if (is.character(data)) {
temp_data <- to_ertmpt_data(read.table(file = data, header = TRUE), model)
data_frame <- temp_data$data
if("transformation" %in% names(temp_data)) {
transformation <- temp_data$transformation
} else transformation <- list()
keep_data_path <- TRUE
} else if (inherits(data, c("ertmpt_data", "rtmpt_data"))) {
data_frame <- data$data
if("transformation" %in% names(data)) {
transformation <- data$transformation
} else transformation <- list()
}
rm(data); gc()
data_elmnts <- c("subj", "group", "tree", "cat", "rt")
if (!all(data_elmnts %in% names(data_frame))) stop("\"data\" must contain \"", data_elmnts[which(!(data_elmnts %in% names(data_frame)))[1]], "\".")
if (!all(data_elmnts == names(data_frame))) {
df <- data_frame
data_frame <- df[,match(data_elmnts, names(data_frame))]
}
if (any(is.na(data_frame)) || min(data_frame) < 0) stop("All values in \"data\" need to be larger than or equal to zero and must not be NA.")
if (max(data_frame[,1]+1) != length(unique(data_frame[,1]))) stop("\"max(data$subj+1)\" must equal \"length(unique(data$subj))\". You might want to use to_ertmpt_data().")
if (max(data_frame[,2]+1) != length(unique(data_frame[,2]))) stop("\"max(data$group+1)\" must equal \"length(unique(data$group))\". You might want to use to_ertmpt_data().")
if (max(data_frame[,3]+1) != length(unique(data_frame[,3]))) stop("\"max(data$tree+1)\" must equal \"length(unique(data$tree))\". You might want to use to_ertmpt_data().")
if (max(data_frame[,4]+1) != length(unique(data_frame[,4]))) stop("\"max(data$cat+1)\" must equal \"length(unique(data$cat))\". You might want to use to_ertmpt_data().")
temp_dir <- gsub("\\\\", "/", tempdir())
data_path <- gsub("\\\\", "/", tempfile(pattern = "ertmpt_data", tmpdir = tempdir(), fileext = ".txt"))
write.table(x = data_frame, file = data_path, sep = " ",
row.names = FALSE, col.names = TRUE)
diag_path <- gsub("\\\\", "/", tempfile(pattern = "diagnosis", tmpdir = tempdir(), fileext = ".out"))
chains_path <- gsub("\\\\", "/", tempfile(pattern = "chains_raw", tmpdir = tempdir(), fileext = ".out"))
loglik_path <- gsub("\\\\", "/", tempfile(pattern = "log_likelihood", tmpdir = tempdir(), fileext = ".out"))
mdl_txt <- gsub("\\\\", "/", tempfile(pattern = "model", tmpdir = tempdir(), fileext = ".txt"))
mdl_info <- gsub("\\\\", "/", tempfile(pattern = "model", tmpdir = tempdir(), fileext = ".info"))
# produce infofile
infofile <- try(get_infofile(model, mdl_txt = mdl_txt, mdl_info = mdl_info))
if(is(infofile, "try-error")) stop("problem with S4 routines in makin infofile.\n")
# # functions to use
# varnames <- function(infofile) {
# varstring <- system(paste("gawk 'END {print}' ", infofile), intern = TRUE)
# var_char <- strsplit(varstring, "\\s+")[[1]]
# return(var_char)
# }
#
# var_char <- varnames(infofile = infofile)
# if (!all(var_char == names(model$params$probs))) stop("Fatal error: Wrong order of parameters in model file.")
# prior parameters
prior_names <- c("mean_of_exp_mu_beta", "var_of_exp_mu_beta", "mean_of_mu_gamma", "var_of_mu_gamma",
"mean_of_omega_sqr", "var_of_omega_sqr", "sf_of_scale_matrix_SIGMA", "sf_of_scale_matrix_GAMMA",
"df_of_sigma_sqr", "prec_epsilon", "add_df_to_invWish")
if (is.null(prior_params)) prior_params <- list()
if (!("mean_of_exp_mu_beta" %in% names(prior_params))) prior_params$mean_of_exp_mu_beta <- 10
if (!("var_of_exp_mu_beta" %in% names(prior_params))) prior_params$var_of_exp_mu_beta <- 100
if (!("mean_of_mu_gamma" %in% names(prior_params))) prior_params$mean_of_mu_gamma <- 0
if (!("var_of_mu_gamma" %in% names(prior_params))) prior_params$var_of_mu_gamma <- 10
if (!("mean_of_omega_sqr" %in% names(prior_params))) prior_params$mean_of_omega_sqr <- 0.005
if (!("var_of_omega_sqr" %in% names(prior_params))) prior_params$var_of_omega_sqr <- 0.01
if (!("df_of_sigma_sqr" %in% names(prior_params))) prior_params$df_of_sigma_sqr <- 2
if (!("sf_of_scale_matrix_SIGMA" %in% names(prior_params))) prior_params$sf_of_scale_matrix_SIGMA <- 1
if (!("sf_of_scale_matrix_GAMMA" %in% names(prior_params))) prior_params$sf_of_scale_matrix_GAMMA <- 1
if (!("prec_epsilon" %in% names(prior_params))) prior_params$prec_epsilon <- 1
if (!("add_df_to_invWish" %in% names(prior_params))) prior_params$add_df_to_invWish <- 1
if (prior_params$add_df_to_invWish %% 1 != 0) stop("\"add_df_to_invWish\" must be an integer!")
if (!all(names(prior_params) %in% prior_names)) {
ind <- which(!(names(prior_params) %in% prior_names))
stop("The following \"prior_params\" are not valid parameters: ", paste0(names(prior_params)[ind], sep = " "))
}
# prior transformation
rate_exp_mu_beta <- prior_params$mean_of_exp_mu_beta / prior_params$var_of_exp_mu_beta
shape_exp_mu_beta <- prior_params$mean_of_exp_mu_beta * rate_exp_mu_beta
rate_omega_sqr <- prior_params$mean_of_omega_sqr / prior_params$var_of_omega_sqr
shape_omega_sqr <- prior_params$mean_of_omega_sqr * rate_omega_sqr
add_df_invWish <- prior_params$add_df_to_invWish - 1
# process names and number of precesses in total
proc_names <- names(model$params$probs)
nprocs <- length(proc_names)
# prepare arguments for RTMPT
CHAR <- c(Data = data_path,
Model = infofile,
Raus = chains_path,
Diagnostics = diag_path,
loglike = loglik_path)
REAL <- c(Rhat_max)
REAL2 <- sapply(X = model$params$probs, FUN = function(x) {ifelse(is.na(x) | x %in% proc_names, -1, x)}) # constants in probs
INTEGER <- c(NoThr = Nchains,
BurnIn = Nwarmup,
Thin = thin,
SampSize = Nchains*Nsamples,
Irep = Irep,
nKERN = dim(model$responses)[1],
nPROCS = nprocs,
nRESP = length(unique(model$responses$MAP)))
INTEGER2 <- model$responses$MAP # cat2resp
BOOL1 <- sapply(X = model$params[["taus"]]["minus",], FUN = function(x) {ifelse(is.na(x) | x %in% proc_names, 1, 0)}) # tau minus
BOOL2 <- sapply(X = model$params[["taus"]]["plus",], FUN = function(x) {ifelse(is.na(x) | x %in% proc_names, 1, 0)}) # tau plus
BOOL3 <- c(loglik = (save_log_lik | indices), bridge = FALSE)#bridge)
REAL3 <- c(prior_params$df_of_sigma_sqr,
shape_omega_sqr,
rate_omega_sqr,
prior_params$mean_of_mu_gamma,
prior_params$var_of_mu_gamma,
prior_params$prec_epsilon,
shape_exp_mu_beta,
rate_exp_mu_beta,
prior_params$sf_of_scale_matrix_SIGMA,
prior_params$sf_of_scale_matrix_GAMMA)
INTEGER3 <- add_df_invWish
F2K <- integer(3*nprocs)
for(i in 1:nprocs) {
theta <- model$params$probs[i]
F2K[i] <- ifelse(theta %in% proc_names, which(proc_names==as.character(theta)),
ifelse(is.na(theta), i, -1))
tauminus <- model$params[["taus"]]["minus", i]
F2K[i+nprocs] <- ifelse(tauminus %in% proc_names, nprocs + which(proc_names==as.character(tauminus)),
ifelse(BOOL1[i], nprocs + i, -1))
tauplus <- model$params[["taus"]]["plus", i]
F2K[i+2*nprocs] <- ifelse(tauplus %in% proc_names, 2*nprocs + which(proc_names==as.character(tauplus)),
ifelse(BOOL2[i], 2*nprocs + i, -1))
}
INTEGER5 <- unique(F2K[F2K != -1])-1
K2F <- integer(3*nprocs)
cntk2f <- 1
for(i in 1:(3*nprocs)) {
tmp <- NULL
if (i <= nprocs) {
tmp <- model$params$probs[i]
} else if (i <= 2*nprocs) {
tmp <- model$params[["taus"]]["minus", i-nprocs]
} else {tmp <- model$params[["taus"]]["plus", i-2*nprocs]}
if(is.na(tmp)) {
K2F[i] <- cntk2f
cntk2f <- cntk2f + 1
} else if(tmp == 0 | (tmp < 1 & tmp > 0)) {
K2F[i] <- -1
} else if(tmp %in% proc_names) {
K2F[i] <- K2F[which(proc_names == as.character(tmp)) + (i>nprocs)*nprocs + (i>2*nprocs)*nprocs]
}
}
INTEGER4 <- K2F-1*(K2F!=-1)
# call C++ function RTMPT
out <- .Call("ertmpt_fit",
as.numeric(REAL),
as.numeric(REAL2),
as.numeric(REAL3),
as.character(CHAR),
as.integer(INTEGER),
as.integer(INTEGER2),
as.integer(INTEGER3),
as.integer(INTEGER4),
as.integer(INTEGER5),
as.logical(BOOL1),
as.logical(BOOL2),
as.logical(BOOL3))
# get data information
data_info <- list(Nsubj = length(unique(data_frame[,1])),
Ngroups = length(unique(data_frame[,2])),
Nprobs = sum(is.na(model$params$probs)),
Nminus = sum(is.na(model$params$taus[1,])),
Nplus = sum(is.na(model$params$taus[2,])),
Nresps = length(unique(model$responses$MAP)),
# epsilon = 1,
probs_string = names(model$params$probs[which(is.na(model$params$probs))]),
minus_string = names(model$params$probs[which(is.na(model$params$taus[1,]))]),
plus_string = names(model$params$probs[which(is.na(model$params$taus[2,]))]),
transformation = transformation)
# prepare output list
ertmpt <- list()
ertmpt$samples <- make_mcmc_list(file = out$pars_samples, infofile = infofile,
Nchains = Nchains, Nsamples = Nsamples,
data_info = data_info, keep = old_label)
out$pars_samples <- NULL; gc()
# diagnostics
ertmpt$diags <- get_diags(diag_file = diag_path, data_info = data_info, keep = old_label)
if (file.exists(diag_path)) file.remove(diag_path)
ertmpt$diags$R_hat <- gelman.diag(ertmpt$samples, multivariate = FALSE)
# specs
infos <- readinfofile(infofile)
ertmpt$specs <- list(model = model, n.chains = Nchains, n.iter = Nsamples, n.burnin = Nwarmup, n.thin = thin,
n.groups = data_info$Ngroups, n.subj = data_info$Nsubj, Irep = Irep, Rhat_max = Rhat_max,
prior_params = prior_params, infolist = infos, call = match.call(),
n.groups = max(data_frame[,2]+1))
if(exists("transformation")) {
ertmpt$specs$transformation <- transformation
}
# WAIC & LOO
if (all(out$LogLik == 0)) {
out$LogLik <- NULL
} else {
if (indices) {
suppressWarnings( temp <- get_indices_x(out$LogLik, Nchains, Nsamples, data_frame) )
ertmpt$indices <- list(WAIC = temp$WAIC, LOO = temp$LOO)
#if (save_log_lik) ertmpt$LogLik <- temp$LogLik
rm(temp)
}
if (save_log_lik) {
ertmpt$LogLik <- out$LogLik
}
}
# if (file.exists(loglik_path)) {
# if (indices) {
# suppressWarnings( temp <- get_indices_x(loglik_path, Nchains, Nsamples, data_frame) )
# ertmpt$indices <- list(WAIC = temp$WAIC, LOO = temp$LOO)
# if (save_log_lik) ertmpt$LogLik <- temp$LogLik
# rm(temp)
# } else {
# if (save_log_lik) {
# fwrite(data.table(V1 = ""), loglik_path, append = TRUE, col.names = FALSE, row.names = FALSE)
# ll_dt <- fread(file = loglik_path, skip = 0, header = FALSE)
# ertmpt$LogLik <- as.matrix(copy(ll_dt))
# }
# }
#
# }
# file.remove(loglik_path)
# remove data file
if (!keep_data_path && file.exists(data_path)) file.remove(data_path)
# remove info file and chains
if (file.exists(infofile)) file.remove(infofile)
if (file.exists(chains_path)) file.remove(chains_path)
# summary
ertmpt$summary <- writeSummaryERTMPT(ertmpt, keep = old_label)
# output
class(ertmpt) <- "ertmpt_fit"
gc()
return(ertmpt)
}
#' Fit Diffusion-RT-MPT Models
#'
#' Given model and data, this function a Hamiltonian MCMC sampler and stores the samples in an mcmc.list called \code{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 \code{diags}. Optionally, the \code{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 (\code{LogLik}) can also be stored.
#' Some specifications of the function call are also saved in \code{specs}.
#'
#' @param model A list of the class \code{drtmpt_model}.
#' @param data Optimally, a list of class \code{drtmpt_data}. Also possible is a \code{data.frame} or a
#' path to the text file. Both, \code{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 \code{\link{to_drtmpt_data}} first, which gives back a
#' \code{drtmpt_data} list with information about the changes in the data, that were needed.
#' @param n.chains Number of chains to use. Default is 4. Must be larger than 1 and smaller or equal to 16.
#' @param n.iter Number of samples per chain. Default is 1000.
#' @param n.phase1 Number of samples for phase 1 (adaptation phase). Default is 1000.
#' @param n.phase2 Number of samples for phase 2. Default is 2000.
#' @param n.thin Thinning factor. Default is 1.
#' @param Rhat_max Maximal Potential scale reduction factor: A lower threshold that needs to be reached before the actual sampling starts. Default is 1.05
#' @param Irep Every \code{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 \code{Irep}:
#' \itemize{
#' \item \code{n.phase1} is a multiple of \code{Irep}
#' \item \code{n.phase2} is a multiple of \code{Irep}
#' \item \code{n.phase1} is smaller than or equal to \code{n.phase2},
#' \item \code{Irep} is a multiple of \code{n.thin} and
#' \item \code{n.iter} is a multiple of \code{Irep / n.thin}.
#' }
#' @param prior_params Named list with prior parameters. All parameters have default values, that lead to uninformative priors. Vectors are not allowed.
#' Allowed parameters are:
#' \itemize{
#' \item \code{prec_epsilon}: prior precision for population means of process-related parameters.
#' Default is \code{1.0}.
#' \item \code{delta_df}: degrees of freedom of t-distribution for motor times. Default is \code{10}.
#' \item \code{delta_mu}: mean of t-distribution for motor times. Default is \code{0.5}.
#' \item \code{delta_scale}: scale of t-distribution for motor times. Default is \code{1.0}.
#' \item \code{SIGMA_Corr_eta}: shape parameter for LKJ distribution for process-related parameters. Default is \code{4.0}.
#' \item \code{SIGMA_SD_rho}: scale parameter of half-Cauchy distribution for process-related parameters. Default is \code{2.5}.
#' \item \code{GAMMA_Corr_eta}: shape parameter for LKJ distribution for motor-related parameters. Default is \code{4.0}.
#' \item \code{GAMMA_SD_rho}: scale parameter of half-Cauchy distribution for motor-related parameters. Default is \code{0.5}.
#' \item \code{Omega2_alpha}: shape parameter of gamma distribution for residual variance. Default is \code{0.0025}.
#' \item \code{Omega2_beta}: rate parameter of gamma distribution for residual variance. Default is \code{0.5}.
#' }
#' @param flags Either NULL or a list of
#' \itemize{
#' \item \code{old_label} If set to \code{TRUE} the old labels of "subj" and "group" of the data will be used in the elements of the output list.
#' Default is \code{FALSE}.
#' \item \code{indices} Model selection indices. If set to \code{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 \code{loo} package. If you want to have the
#' log-likelihood matrix stored in the output of this function, you can set \code{loglik} to \code{TRUE}. Default for
#' \code{indices} is \code{FALSE}.
#' \item \code{loglik} If set to \code{TRUE} and \code{indices = TRUE} the log-likelihood matrix for each iteration and trial will
#' be saved in the output as a matrix. Default is \code{FALSE}.
#' \item \code{random_init} If set to \code{TRUE} the initial values are randomly drawn. If \code{FLASE} maximum likelihood is used
#' for initial values.
#' }
#' @param control Either NULL or a list of
#' \itemize{
#' \item \code{maxthreads} for the ML estimation of the initial values and the calculation of the DIC values one can use more than
#' \code{n.chains} threads for parallelization. Default is 4 like \code{n.chians}. \code{maxthreads} must be larger or equal to
#' \code{n.chains}.
#' \item \code{maxtreedepth1_3} maxtree-depth of the no-U-turn algorithm in Phases 1 to 3
#' \item \code{maxtreedepth4} maxtree-depth of the no-U-turn algorithm in Phases 4
#' }
#' @return A list of the class \code{drtmpt_fit} containing
#' \itemize{
#' \item \code{samples}: the posterior samples as an \code{mcmc.list} object,
#' \item \code{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,
#' \item \code{specs}: some model specifications like the model, arguments of the model call, and information about the data transformation,
#' \item \code{indices} (optional): if enabled, WAIC and LOO,
#' \item \code{LogLik} (optional): if enabled, the log-likelihood matrix used for WAIC and LOO.
#' \item \code{summary} includes posterior mean and median of the main parameters.
#' }
#' @references
#' Klauer, K. C. (2010). Hierarchical multinomial processing tree models: A latent-trait approach. \emph{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.
#' \emph{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.
#' \emph{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.
#' \emph{Journal of Machine Learning Research, 11(Dec)}, 3571-3594.
#'
#' @examples
#' ####################################################################################
#' # 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
#' }
#' @author Raphael Hartmann
#' @useDynLib "rtmpt", .registration=TRUE
#' @export
#' @importFrom coda gelman.diag
#' @importFrom methods as callGeneric new is
#' @importFrom utils read.table write.table
fit_drtmpt <- function(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) {
# CHECKS
model_elmnts <- c("lines", "params", "responses")
if (!is.list(model)) stop("\"model\" must be a list.")
if (!all(model_elmnts %in% names(model))) stop("\"model\" must contain \"", model_elmnts[which(!(model_elmnts %in% names(model)))[1]], "\".")
if (n.chains < 2 || n.chains > 16) stop("\"n.chains\" must be larger than 1 and lower or equal to 16.")
if (Irep < 1) stop("\"Irep\" must be larger than or equal to one.")
if (n.thin < 1) stop("\"n.thin\" must be larger than or equal to one.")
if (n.phase1 %% Irep) stop("\"n.phase1\" must be a multiple of \"Irep\".")
if (n.phase2 %% Irep) stop("\"n.phase2\" must be a multiple of \"Irep\".")
if (n.phase1 > n.phase2) stop("\"n.phase2\" must be larger or equal to \"n.phase1\".")
if (Irep %% n.thin != 0) stop("\"Irep\" must be a multiple of \"n.thin\".")
if (n.iter %% (Irep/n.thin) != 0) stop("\"n.iter\" must be a multiple of \"Irep\" / \"n.thin\".")
if (n.iter < Irep/n.thin) stop("\"n.iter\" must be greater or equal to \"Irep\" / \"n.thin\" = ", Irep*n.thin, ".")
if (Rhat_max < 1) stop("\"Rhat_max\" must be larger than or equal to one.")
if (!is.null(prior_params) && !is.list(prior_params)) stop("\"prior_params\" must be a list or NULL.")
if (!is.null(flags) && !is.list(flags)) stop("\"flags\" must be a list or NULL.")
if (!is.null(control) && !is.list(control)) stop("\"control\" must be a list or NULL.")
if (!is.null(flags)) {
if ("old_label" %in% names(flags)) if (!is.logical(flags$old_label)) stop("\"flags$old_label\" must either be TRUE or FALSE.")
if ("indices" %in% names(flags)) if (!is.logical(flags$indices)) stop("\"flags$indices\" must either be TRUE or FALSE.")
if ("loglik" %in% names(flags)) if (!is.logical(flags$loglik)) stop("\"flags$loglik\" must either be TRUE or FALSE.")
if ("random_init" %in% names(flags)) if (!is.logical(flags$random_init)) stop("\"flags$random_init\" must either be TRUE or FALSE.")
}
if (!is.null(control)) {
if ("maxthreads" %in% control) if (!is.numeric(control$maxthreads)) stop("\"control$maxthreads\" must be numeric.")
if ("maxtreedepth1_3" %in% control) if (!is.numeric(control$maxtreedepth1_3)) stop("\"control$maxtreedepth1_3\" must be numeric.")
if ("maxtreedepth4" %in% control) if (!is.numeric(control$maxtreedepth4)) stop("\"control$maxtreedepth4\" must be numeric.")
}
# SET FLAGS
if (!is.list(flags)) flags <- list()
if (!"old_label" %in% names(flags)) {
old_label <- FALSE
} else {
old_label <- flags$old_label
}
if (!"indices" %in% names(flags)) flags$indices <- FALSE
if (!"loglik" %in% names(flags)) flags$loglik <- FALSE
if (!"random_init" %in% names(flags)) flags$random_init <- FALSE
# SET CONTROLS
if (!is.list(control)) control <- list()
if (!"maxthreads" %in% names(flags)) control$maxthreads <- 4
if (!"maxtreedepth1_3" %in% names(flags)) control$maxtreedepth1_3 <- 5
if (!"maxtreedepth4" %in% names(flags)) control$maxtreedepth4 <- 9
# PREPARE DATA
keep_data_path <- FALSE
data_frame <- NULL
if (is.data.frame(data)) {
temp_data <- to_drtmpt_data(data, model)
data_frame <- temp_data$data
if("transformation" %in% names(temp_data)) {
transformation <- temp_data$transformation
} else transformation <- list()
} else if (is.character(data)) {
temp_data <- to_drtmpt_data(read.table(file = data, header = TRUE), model)
data_frame <- temp_data$data
if("transformation" %in% names(temp_data)) {
transformation <- temp_data$transformation
} else transformation <- list()
keep_data_path <- TRUE
} else if (inherits(data, "drtmpt_data")) {
data_frame <- data$data
if("transformation" %in% names(data)) {
transformation <- data$transformation
} else transformation <- list()
}
rm(data); gc()
data_elmnts <- c("subj", "group", "tree", "cat", "rt")
if (!all(data_elmnts %in% names(data_frame))) stop("\"data\" must contain \"", data_elmnts[which(!(data_elmnts %in% names(data_frame)))[1]], "\".")
if (!all(data_elmnts == names(data_frame))) {
df <- data_frame
data_frame <- df[,match(data_elmnts, names(data_frame))]
}
if (any(is.na(data_frame)) || min(data_frame) < 0) stop("All values in \"data\" need to be larger than or equal to zero and must not be NA.")
if (max(data_frame[,1]+1) != length(unique(data_frame[,1]))) stop("\"max(data$subj+1)\" must equal \"length(unique(data$subj))\". You might want to use to_rtmpt_data().")
if (max(data_frame[,2]+1) != length(unique(data_frame[,2]))) stop("\"max(data$group+1)\" must equal \"length(unique(data$group))\". You might want to use to_rtmpt_data().")
if (max(data_frame[,3]+1) != length(unique(data_frame[,3]))) stop("\"max(data$tree+1)\" must equal \"length(unique(data$tree))\". You might want to use to_rtmpt_data().")
if (max(data_frame[,4]+1) != length(unique(data_frame[,4]))) stop("\"max(data$cat+1)\" must equal \"length(unique(data$cat))\". You might want to use to_rtmpt_data().")
# PREPARE PATHS
# Inputs
model_path <- gsub("\\\\", "/", tempfile(pattern = "model", tmpdir = tempdir(), fileext = ".info"))
mdl_path <- gsub("\\\\", "/", tempfile(pattern = "model", tmpdir = tempdir(), fileext = ".txt"))
data_path <- gsub("\\\\", "/", tempfile(pattern = "data", tmpdir = tempdir(), fileext = ".txt"))
write.table(x = data_frame, file = data_path, sep = " ", row.names = FALSE, col.names = TRUE)
# Outputs
raus_path <- gsub("\\\\", "/", tempfile(pattern = "raus", tmpdir = tempdir(), fileext = ".out"))
ll_path <- gsub("\\\\", "/", tempfile(pattern = "loglik", tmpdir = tempdir(), fileext = ".out"))
cont_path <- gsub("\\\\", "/", tempfile(pattern = "continue", tmpdir = tempdir(), fileext = ".out"))
means_path <- gsub("\\\\", "/", tempfile(pattern = "means", tmpdir = tempdir(), fileext = ".out"))
tests_path <- gsub("\\\\", "/", tempfile(pattern = "tests", tmpdir = tempdir(), fileext = ".out"))
rand_path <- gsub("\\\\", "/", tempfile(pattern = "random", tmpdir = tempdir(), fileext = ".out"))
# PREPARE INFOFILE
infofile <- try(get_infofile(model, mdl_txt = mdl_path, mdl_info = model_path))
if(is(infofile, "try-error")) stop("problem with S4 routines in makin infofile.\n")
# Prior Parameter
prior_names <- c("prec_epsilon", "delta_df", "delta_mu", "delta_scale",
"SIGMA_Corr_eta", "SIGMA_SD_rho", "GAMMA_Corr_eta", "GAMMA_SD_rho",
"Omega2_alpha", "Omega2_beta")
if (is.null(prior_params)) prior_params <- list()
if (!("prec_epsilon" %in% names(prior_params))) prior_params$prec_epsilon <- 1.0
if (!("delta_df" %in% names(prior_params))) prior_params$delta_df <- 10
if (!("delta_mu" %in% names(prior_params))) prior_params$delta_mu <- 0.5
if (!("delta_scale" %in% names(prior_params))) prior_params$delta_scale <- 1.0
if (!("SIGMA_Corr_eta" %in% names(prior_params))) prior_params$SIGMA_Corr_eta <- 4.0
if (!("SIGMA_SD_rho" %in% names(prior_params))) prior_params$SIGMA_SD_rho <- 2.5
if (!("GAMMA_Corr_eta" %in% names(prior_params))) prior_params$GAMMA_Corr_eta <- 4.0
if (!("GAMMA_SD_rho" %in% names(prior_params))) prior_params$GAMMA_SD_rho <- 0.5
if (!("Omega2_alpha" %in% names(prior_params))) prior_params$Omega2_alpha <- 0.0025
if (!("Omega2_beta" %in% names(prior_params))) prior_params$Omega2_beta <- 0.5
if (!is.numeric(unlist(prior_params))) stop("prior_params must be a list of numerics and integers")
if (prior_params$delta_df %% 1 != 0) stop("delta_df must be an integer")
# PROCESS NAMES AND NUMBER
proc_names <- names(model$params$threshold)
nprocs <- length(proc_names)
# KERN2FREE
K2F <- integer(3*nprocs)
cntk2f <- integer(3)
consts <- rep(NaN, 3*nprocs)
comp <- rep(FALSE, 3*nprocs)
cntcomp <- integer(3)
for (i in 1:3) {
cntk2f[i] <- 0
for (j in 1:nprocs) {
tmp <- NULL
if (i == 1) {
tmp <- model$params$threshold[1, j]
} else if (i == 2) {
tmp <- model$params$driftrate[1, j]
} else if (i == 3) {
tmp <- model$params$startpoint[1, j]
}
ind <- (i-1)*nprocs+j
if(is.na(tmp) | is.numeric(tmp)) {
K2F[ind] <- cntk2f[i]
cntk2f[i] <- cntk2f[i] + 1
if (is.na(tmp)) {
comp[(cntk2f[i]-1)*3 + i] <- TRUE
}
if (is.numeric(tmp)) {
consts[(cntk2f[i]-1)*3 + i] <- tmp
}
} else if(tmp %in% proc_names) {
K2F[ind] <- K2F[which(proc_names == as.character(tmp)) + (i>=2)*nprocs + (i==3)*nprocs]
}
}
}
# PATHS ARGUMENT
CHAR1 <- as.character(c(Data = data_path,
Model = infofile,
Raus = raus_path,
LogLik = ll_path,
Continue = cont_path,
Means = means_path,
Tests = tests_path,
Random = rand_path,
tmpdir = paste0(tempdir(), "/")))
# MCMC SETTINGS ARGUMENT
INTEGER1 <- as.integer(c(IREP = Irep,
PHASE1 = n.phase1,
PHASE2 = n.phase2,
THIN = n.thin,
NOTHREADS = n.chains,
SAMPLE_SIZE = n.chains*n.iter,
MAXTHREADS = control$maxthreads,
nKERN = dim(model$responses)[1],
nPROCS = nprocs,
respno = length(unique(model$responses$MAP)),
cat2resp = model$responses$MAP))
REAL1 <- c(Rhat_max = Rhat_max)
# FLAGS ARGUMENT
BOOL1 <- as.integer(c(DIC_FL = flags$indices,
LOG_LIK_FL = (flags$loglik),
ML_INIT_FL = !flags$random_init))
# PRIORS ARGUMENT
INTEGER2 <- as.integer(c(degf = prior_params$delta_df))
REAL2 <- c(PRIOR = prior_params$prec_epsilon,
etat = prior_params$SIGMA_Corr_eta,
taut = prior_params$SIGMA_SD_rho,
etar = prior_params$GAMMA_Corr_eta,
taur = prior_params$GAMMA_SD_rho,
mu_prior = prior_params$delta_mu,
rsd = prior_params$delta_scale,
prioralpha = prior_params$Omega2_alpha,
priorbeta = prior_params$Omega2_beta)
# HMC OPTIONS ARGUMENT
INTEGER3 <- as.integer(c(maxtreedepth1_3 = control$maxtreedepth1_3,
maxtreedepth4 = control$maxtreedepth4))
# CONTINUATION ARGUMENT
INTEGER4 <- as.integer(c(goon = FALSE,
ADDITION = 0))
# CONSTANTS AND EQUALIZATION
INTEGER5 <- as.integer(c(kern2free = K2F,
comp = comp,
ifree = cntk2f))
REAL3 <- c(consts = consts)
# C++ FUNCTION DRTMPT CALL
out <- .Call("drtmpt_fit", CHAR1, INTEGER1, REAL1, BOOL1,
INTEGER2, REAL2,
INTEGER3, INTEGER4,
REAL3, INTEGER5)
file.remove(data_path)
# DATA INFORMATION
data_info <- list(Nsubj = length(unique(data_frame[,1])),
Ngroups = length(unique(data_frame[,2])),
Nthreshold = sum(is.na(model$params$threshold[1,])),
Ndriftrate = sum(is.na(model$params$driftrate[1,])),
Nstartpoint = sum(is.na(model$params$startpoint[1,])),
Nresps = length(unique(model$responses$MAP)),
thresh_string = names(model$params$threshold[which(is.na(model$params$threshold[1,]))]),
drift_string = names(model$params$driftrate[which(is.na(model$params$driftrate[1,]))]),
start_string = names(model$params$startpoint[which(is.na(model$params$startpoint[1,]))]),
transformation = transformation)
# PREPARE OUTPUT LIST
drtmpt <- list()
drtmpt$samples <- make_mcmc_list_d(file = out$pars_samples, infofile = infofile,
Nchains = n.chains, Nsamples = n.iter,
data_info = data_info, keep = old_label)
file.remove(raus_path)
# DIAGNOSTICS
drtmpt$diags <- get_diags_d(diag_file = tests_path, data_info = data_info, keep = old_label, DIC = flags$indices)
drtmpt$diags$R_hat <- gelman.diag(drtmpt$samples, multivariate = FALSE)
# SPECS
infos <- readinfofile(infofile)
drtmpt$specs <- list(model = model, n.chains = n.chains, n.iter = n.iter, n.phase1 = n.phase1, n.phase2 = n.phase2,
n.thin = n.thin, n.groups = data_info$Ngroups, n.subj = data_info$Nsubj, Irep = Irep,
Rhat_max = Rhat_max, prior_params = prior_params, infolist = infos, call = match.call())
if(exists("transformation")) {
drtmpt$specs$transformation <- transformation
}
# # COMPUTE WAIC AND LOO-IC
# if (flags$indices) {
# if (file.exists(ll_path)) {
# if (flags$indices) {
# temp <- NULL
# suppressWarnings( temp <- get_indices_x(log_lik = ll_path, Nchains = n.chains, Nsample = n.iter, df = data_frame) )
# drtmpt$indices <- list(WAIC = temp$WAIC, LOO = temp$LOO)
# if (flags$loglik) drtmpt$LogLik <- temp$LogLik
# rm(temp)
# } else {
# if (flags$loglik) {
# fwrite(data.table(V1 = ""), ll_path, append = TRUE, col.names = FALSE, row.names = FALSE)
# ll_dt <- fread(file = ll_path, skip = 0, header = FALSE)
# drtmpt$LogLik <- as.matrix(copy(ll_dt))
# }
# }
# file.remove(ll_path)
# }
# }
# WAIC & LOO
if (all(out$loglik == 0)) {
out$loglik <- NULL
} else {
if (flags$indices) {
suppressWarnings( temp <- get_indices_x(out$loglik, n.chains, n.iter, data_frame) )
drtmpt$indices <- list(WAIC = temp$WAIC, LOO = temp$LOO)
rm(temp)
}
if (flags$loglik) {
drtmpt$loglik <- out$loglik
}
}
# CONTINUATION
if (FALSE) {
drtmpt$continue <- readLines(con = cont_path)
}
file.remove(cont_path)
# SUMMARY
drtmpt$summary <- writeSummaryDRTMPT(x = drtmpt, keep = old_label)
# CLEAN-UP
file.remove(means_path)
file.remove(tests_path)
file.remove(rand_path)
file.remove(infofile)
file.remove(mdl_path)
# OUTPUT
class(drtmpt) <- "drtmpt_fit"
gc()
return(drtmpt)
}
Try the rtmpt package in your browser
Any scripts or data that you put into this service are public.
rtmpt documentation built on April 4, 2025, 12:37 a.m.
R Package Documentation
Browse R Packages
We want your feedback!
Note that we can't provide technical support on individual packages. You should contact the package authors for that.
Defines functions fit_drtmpt fit_ertmpt
Documented in fit_drtmpt fit_ertmpt
#' Fit Exponential-RT-MPT Models
#'
#' Given model and data, this function calls an altered version of the C++ program by Klauer and Kellen (2018) to sample from
#' the posterior distribution via a Metropolis-Gibbs sampler and storing it in an mcmc.list called \code{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 \code{diags}. Optionally, the \code{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 (\code{LogLik}) can also be stored.
#' Some specifications of the function call are also saved in \code{specs}.
#'
#' @param model A list of the class \code{ertmpt_model}.
#' @param data Optimally, a list of class \code{ertmpt_data}. Also possible is a \code{data.frame} or a
#' path to the text file. Both, \code{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 \code{\link{to_ertmpt_data}} first, which gives back an \code{ertmpt_data} list
#' with informations about the changes in the data, that were needed.
#' @param n.chains Number of chains to use. Default is 4. Must be larger than 1 and smaller or equal to 16.
#' @param n.iter Number of samples per chain. Default is 5000.
#' @param n.burnin Number of warm-up samples. Default is 200.
#' @param n.thin Thinning factor. Default is 1.
#' @param Rhat_max Maximal Potential scale reduction factor: A lower threshold that needs to be reached before the actual sampling starts. Default is 1.05
#' @param Irep Every \code{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 \code{Irep}:
#' \itemize{
#' \item \code{n.burnin} is smaller than or equal to \code{Irep},
#' \item \code{Irep} is a multiple of \code{n.thin} and
#' \item \code{n.iter} is a multiple of \code{Irep / n.thin}.
#' }
#' @param prior_params Named list with prior parameters. All parameters have default values, that lead to uninformative priors. Vectors are not allowed.
#' Allowed parameters are:
#' \itemize{
#' \item \code{mean_of_exp_mu_beta}: This is the a priori expected exponential rate (\code{E(exp(beta)) = E(lambda)}) and
#' \code{1/mean_of_exp_mu_beta} is the a priori expected process time (\code{1/E(exp(beta)) = E(tau)}). The default
#' mean is set to \code{10}, such that the expected a priori process time is \code{0.1} seconds.
#' \item \code{var_of_exp_mu_beta}: The a priori group-specific variance of the exponential rates. Since
#' \code{exp(mu_beta)} is Gamma distributed, the rate of the distribution is just mean divided by variance and
#' the shape is the mean times the rate. The default is set to \code{100}.
#' \item \code{mean_of_mu_gamma}: This is the a priori expected \emph{mean parameter} of the encoding and response execution times,
#' which follow a normal distribution truncated from below at zero, so \code{E(mu_gamma) < E(gamma)}. The default is \code{0}.
#' \item \code{var_of_mu_gamma}: The a priori group-specific variance of the \emph{mean parameter}. Its default is \code{10}.
#' \item \code{mean_of_omega_sqr}: This is the a priori expected residual variance (\code{E(omega^2)}). Its distribution
#' differs from the one used in the paper. Here it is a Gamma distribution instead of an improper one. The default
#' is \code{0.005}.
#' \item \code{var_of_omega_sqr}: The a priori variance of the residual variance (\code{Var(omega^2)}). The default is
#' \code{0.01}. The default of the mean and variance is equivalent to a shape and rate of \code{0.0025} and
#' \code{0.5}, respectivly.
#' \item \code{df_of_sigma_sqr}: A priori degrees of freedom for the individual variance of the response executions. The
#' individual variance has a scaled inverse chi-squared prior with \code{df_of_sigma_sqr} degrees of freedom and
#' \code{omega^2} as scale. \code{2} is the default and it should be an integer.
#' \item \code{sf_of_scale_matrix_SIGMA}: The original scaling matrix (S) of the (scaled) inverse Wishart distribution for the process
#' related parameters is an identity matrix \code{S=I}. \code{sf_of_scale_matrix_SIGMA} is a scaling factor, that scales this
#' matrix (\code{S=sf_of_scale_matrix_SIGMA*I}). Its default is \code{1}.
#' \item \code{sf_of_scale_matrix_GAMMA}: The original scaling matrix (S) of the (scaled) inverse Wishart distribution for the encoding and
#' motor execution parameters is an identity matrix \code{S=I}. \code{sf_of_scale_matrix_GAMMA} is a scaling factor, that scales
#' this matrix (\code{S=sf_of_scale_matrix_GAMMA*I}). Its default is \code{1}.
#' \item \code{prec_epsilon}: This is epsilon in the paper. It is the precision of mu_alpha and all xi (scaling parameter
#' in the scaled inverse Wishart distribution). Its default is also \code{1}.
#' \item \code{add_df_to_invWish}: If \code{P} is the number of parameters or rather the size of the scale matrix used in the (scaled)
#' inverse Wishart distribution then \code{add_df_to_invWish} is the number of degrees of freedom that can be added to it. So
#' \code{DF = P + add_df_to_invWish}. The default for \code{add_df_to_invWish} is \code{1}, such that the correlations are uniformly
#' distributed within \code{[-1, 1]}.
#' }
#' @param indices Model selection indices. If set to \code{TRUE} the log-likelihood for each iteration and trial will be stored temporarily
#' and with that the WAIC and LOO will be calculated via the \code{loo} package. If you want to have this log-likelihood matrix stored in the
#' output of this function, you can set \code{save_log_lik} to \code{TRUE}. The default for \code{indices} is \code{FALSE}.
#' @param save_log_lik If set to \code{TRUE} the log-likelihood matrix for each iteration and trial will
#' be saved in the output as a matrix. Its default is \code{FALSE}.
#' @param old_label If set to \code{TRUE} the old labels of "subj" and "group" of the data will be used in the elements of the output list. Default is \code{FALSE}.
#' @return A list of the class \code{ertmpt_fit} containing
#' \itemize{
#' \item \code{samples}: the posterior samples as an \code{mcmc.list} object,
#' \item \code{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,
#' \item \code{specs}: some model specifications like the model, arguments of the model call, and information about the data transformation,
#' \item \code{indices} (optional): if enabled, WAIC and LOO,
#' \item \code{LogLik} (optional): if enabled, the log-likelihood matrix used for WAIC and LOO.
#' \item \code{summary} includes posterior mean and median of the main parameters.
#' }
#' @references
#' Hartmann, R., Johannsen, L., & Klauer, K. C. (2020). rtmpt: An R package for fitting response-time extended multinomial processing tree models.
#' \emph{Behavior Research Methods, 52}(3), 1313–1338.
#'
#' Hartmann, R., & Klauer, K. C. (2020). Extending RT-MPTs to enable equal process times. \emph{Journal of Mathematical Psychology, 96}, 102340.
#'
#' Klauer, K. C. (2010). Hierarchical multinomial processing tree models: A latent-trait approach. \emph{Psychometrika, 75(1)}, 70-98.
#'
#' Klauer, K. C., & Kellen, D. (2018). RT-MPTs: Process models for response-time distributions based on multinomial processing trees with
#' applications to recognition memory. \emph{Journal of Mathematical Psychology, 82}, 111-130.
#'
#' Spiegelhalter, D. J., Best, N. G., Carlin, B. P., & Van Der Linde, A. (2002). Bayesian measures of model complexity and fit.
#' \emph{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.
#' \emph{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.
#' \emph{Journal of Machine Learning Research, 11(Dec)}, 3571-3594.
#' @examples
#' ####################################################################################
#' # 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_ertmpt_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_ertmpt_data(raw_data = data, model = model)
#' \donttest{
#' # This might take some time
#' ertmpt_out <- fit_ertmpt(model = model, data = data_list, Rhat_max = 1.1)
#' ertmpt_out
#' }
#' # Type ?SimData for another working example.
#' @author Raphael Hartmann
#' @useDynLib "rtmpt", .registration=TRUE
#' @export
#' @importFrom coda gelman.diag
#' @importFrom methods as callGeneric new is
#' @importFrom utils read.table write.table
fit_ertmpt <- function(model,
data,
n.chains = 4,
n.iter = 5000,
n.burnin = 200,
n.thin = 1,
Rhat_max = 1.05,
Irep = 1000,
prior_params = NULL,
indices = FALSE,
save_log_lik = FALSE,
old_label = FALSE) {
# internal argument name changes
Nchains <- n.chains
Nsamples <- n.iter
Nwarmup <- n.burnin
thin <- n.thin
# some controls
model_elmnts <- c("lines", "params", "responses")
if (!is.list(model)) stop("\"model\" must be a list.")
if (!all(model_elmnts %in% names(model))) stop("\"model\" must contain \"", model_elmnts[which(!(model_elmnts %in% names(model)))[1]], "\".")
if (Nchains < 2 || Nchains > 16) stop("\"n.chains\" must be larger than 1 and lower or equal to 16.")
if (Irep < 1) stop("\"Irep\" must be larger than or equal to one.")
if (thin < 1) stop("\"n.thin\" must be larger than or equal to one.")
if (Irep %% thin != 0) stop("\"Irep\" must be a multiple of \"n.thin\".")
if (Nsamples %% (Irep/thin) != 0) stop("\"n.iter\" must be a multiple of \"Irep\" / \"n.thin\".")
if (Nsamples < Irep/thin) stop("\"n.iter\" must be greater or equal to \"Irep\" / \"n.thin\" = ", Irep*thin, ".")
if (Nwarmup > Irep) stop("\"n.burnin\" must be smaller than or equal to \"Irep\".")
if (Rhat_max < 1) stop("\"Rhat_max\" must be larger than or equal to one.")
if (!is.logical(indices)) stop("\"indices\" must either be TRUE or FALSE.")
# if (!is.logical(bridge)) stop("\"bridge\" must either be TRUE or FALSE.")
bridge = FALSE
if (!is.logical(save_log_lik)) stop("\"save_log_lik\" must either be TRUE or FALSE.")
if (!is.logical(old_label)) stop("\"old_label\" must either be TRUE or FALSE.")
if (!is.null(prior_params) && !is.list(prior_params)) stop("\"prior_params\" must be a list.")
# prepare data
keep_data_path <- FALSE
if (is.data.frame(data)) {
temp_data <- to_ertmpt_data(data, model)
data_frame <- temp_data$data
if("transformation" %in% names(temp_data)) {
transformation <- temp_data$transformation
} else transformation <- list()
} else if (is.character(data)) {
temp_data <- to_ertmpt_data(read.table(file = data, header = TRUE), model)
data_frame <- temp_data$data
if("transformation" %in% names(temp_data)) {
transformation <- temp_data$transformation
} else transformation <- list()
keep_data_path <- TRUE
} else if (inherits(data, c("ertmpt_data", "rtmpt_data"))) {
data_frame <- data$data
if("transformation" %in% names(data)) {
transformation <- data$transformation
} else transformation <- list()
}
rm(data); gc()
data_elmnts <- c("subj", "group", "tree", "cat", "rt")
if (!all(data_elmnts %in% names(data_frame))) stop("\"data\" must contain \"", data_elmnts[which(!(data_elmnts %in% names(data_frame)))[1]], "\".")
if (!all(data_elmnts == names(data_frame))) {
df <- data_frame
data_frame <- df[,match(data_elmnts, names(data_frame))]
}
if (any(is.na(data_frame)) || min(data_frame) < 0) stop("All values in \"data\" need to be larger than or equal to zero and must not be NA.")
if (max(data_frame[,1]+1) != length(unique(data_frame[,1]))) stop("\"max(data$subj+1)\" must equal \"length(unique(data$subj))\". You might want to use to_ertmpt_data().")
if (max(data_frame[,2]+1) != length(unique(data_frame[,2]))) stop("\"max(data$group+1)\" must equal \"length(unique(data$group))\". You might want to use to_ertmpt_data().")
if (max(data_frame[,3]+1) != length(unique(data_frame[,3]))) stop("\"max(data$tree+1)\" must equal \"length(unique(data$tree))\". You might want to use to_ertmpt_data().")
if (max(data_frame[,4]+1) != length(unique(data_frame[,4]))) stop("\"max(data$cat+1)\" must equal \"length(unique(data$cat))\". You might want to use to_ertmpt_data().")
temp_dir <- gsub("\\\\", "/", tempdir())
data_path <- gsub("\\\\", "/", tempfile(pattern = "ertmpt_data", tmpdir = tempdir(), fileext = ".txt"))
write.table(x = data_frame, file = data_path, sep = " ",
row.names = FALSE, col.names = TRUE)
diag_path <- gsub("\\\\", "/", tempfile(pattern = "diagnosis", tmpdir = tempdir(), fileext = ".out"))
chains_path <- gsub("\\\\", "/", tempfile(pattern = "chains_raw", tmpdir = tempdir(), fileext = ".out"))
loglik_path <- gsub("\\\\", "/", tempfile(pattern = "log_likelihood", tmpdir = tempdir(), fileext = ".out"))
mdl_txt <- gsub("\\\\", "/", tempfile(pattern = "model", tmpdir = tempdir(), fileext = ".txt"))
mdl_info <- gsub("\\\\", "/", tempfile(pattern = "model", tmpdir = tempdir(), fileext = ".info"))
# produce infofile
infofile <- try(get_infofile(model, mdl_txt = mdl_txt, mdl_info = mdl_info))
if(is(infofile, "try-error")) stop("problem with S4 routines in makin infofile.\n")
# # functions to use
# varnames <- function(infofile) {
# varstring <- system(paste("gawk 'END {print}' ", infofile), intern = TRUE)
# var_char <- strsplit(varstring, "\\s+")[[1]]
# return(var_char)
# }
#
# var_char <- varnames(infofile = infofile)
# if (!all(var_char == names(model$params$probs))) stop("Fatal error: Wrong order of parameters in model file.")
# prior parameters
prior_names <- c("mean_of_exp_mu_beta", "var_of_exp_mu_beta", "mean_of_mu_gamma", "var_of_mu_gamma",
"mean_of_omega_sqr", "var_of_omega_sqr", "sf_of_scale_matrix_SIGMA", "sf_of_scale_matrix_GAMMA",
"df_of_sigma_sqr", "prec_epsilon", "add_df_to_invWish")
if (is.null(prior_params)) prior_params <- list()
if (!("mean_of_exp_mu_beta" %in% names(prior_params))) prior_params$mean_of_exp_mu_beta <- 10
if (!("var_of_exp_mu_beta" %in% names(prior_params))) prior_params$var_of_exp_mu_beta <- 100
if (!("mean_of_mu_gamma" %in% names(prior_params))) prior_params$mean_of_mu_gamma <- 0
if (!("var_of_mu_gamma" %in% names(prior_params))) prior_params$var_of_mu_gamma <- 10
if (!("mean_of_omega_sqr" %in% names(prior_params))) prior_params$mean_of_omega_sqr <- 0.005
if (!("var_of_omega_sqr" %in% names(prior_params))) prior_params$var_of_omega_sqr <- 0.01
if (!("df_of_sigma_sqr" %in% names(prior_params))) prior_params$df_of_sigma_sqr <- 2
if (!("sf_of_scale_matrix_SIGMA" %in% names(prior_params))) prior_params$sf_of_scale_matrix_SIGMA <- 1
if (!("sf_of_scale_matrix_GAMMA" %in% names(prior_params))) prior_params$sf_of_scale_matrix_GAMMA <- 1
if (!("prec_epsilon" %in% names(prior_params))) prior_params$prec_epsilon <- 1
if (!("add_df_to_invWish" %in% names(prior_params))) prior_params$add_df_to_invWish <- 1
if (prior_params$add_df_to_invWish %% 1 != 0) stop("\"add_df_to_invWish\" must be an integer!")
if (!all(names(prior_params) %in% prior_names)) {
ind <- which(!(names(prior_params) %in% prior_names))
stop("The following \"prior_params\" are not valid parameters: ", paste0(names(prior_params)[ind], sep = " "))
}
# prior transformation
rate_exp_mu_beta <- prior_params$mean_of_exp_mu_beta / prior_params$var_of_exp_mu_beta
shape_exp_mu_beta <- prior_params$mean_of_exp_mu_beta * rate_exp_mu_beta
rate_omega_sqr <- prior_params$mean_of_omega_sqr / prior_params$var_of_omega_sqr
shape_omega_sqr <- prior_params$mean_of_omega_sqr * rate_omega_sqr
add_df_invWish <- prior_params$add_df_to_invWish - 1
# process names and number of precesses in total
proc_names <- names(model$params$probs)
nprocs <- length(proc_names)
# prepare arguments for RTMPT
CHAR <- c(Data = data_path,
Model = infofile,
Raus = chains_path,
Diagnostics = diag_path,
loglike = loglik_path)
REAL <- c(Rhat_max)
REAL2 <- sapply(X = model$params$probs, FUN = function(x) {ifelse(is.na(x) | x %in% proc_names, -1, x)}) # constants in probs
INTEGER <- c(NoThr = Nchains,
BurnIn = Nwarmup,
Thin = thin,
SampSize = Nchains*Nsamples,
Irep = Irep,
nKERN = dim(model$responses)[1],
nPROCS = nprocs,
nRESP = length(unique(model$responses$MAP)))
INTEGER2 <- model$responses$MAP # cat2resp
BOOL1 <- sapply(X = model$params[["taus"]]["minus",], FUN = function(x) {ifelse(is.na(x) | x %in% proc_names, 1, 0)}) # tau minus
BOOL2 <- sapply(X = model$params[["taus"]]["plus",], FUN = function(x) {ifelse(is.na(x) | x %in% proc_names, 1, 0)}) # tau plus
BOOL3 <- c(loglik = (save_log_lik | indices), bridge = FALSE)#bridge)
REAL3 <- c(prior_params$df_of_sigma_sqr,
shape_omega_sqr,
rate_omega_sqr,
prior_params$mean_of_mu_gamma,
prior_params$var_of_mu_gamma,
prior_params$prec_epsilon,
shape_exp_mu_beta,
rate_exp_mu_beta,
prior_params$sf_of_scale_matrix_SIGMA,
prior_params$sf_of_scale_matrix_GAMMA)
INTEGER3 <- add_df_invWish
F2K <- integer(3*nprocs)
for(i in 1:nprocs) {
theta <- model$params$probs[i]
F2K[i] <- ifelse(theta %in% proc_names, which(proc_names==as.character(theta)),
ifelse(is.na(theta), i, -1))
tauminus <- model$params[["taus"]]["minus", i]
F2K[i+nprocs] <- ifelse(tauminus %in% proc_names, nprocs + which(proc_names==as.character(tauminus)),
ifelse(BOOL1[i], nprocs + i, -1))
tauplus <- model$params[["taus"]]["plus", i]
F2K[i+2*nprocs] <- ifelse(tauplus %in% proc_names, 2*nprocs + which(proc_names==as.character(tauplus)),
ifelse(BOOL2[i], 2*nprocs + i, -1))
}
INTEGER5 <- unique(F2K[F2K != -1])-1
K2F <- integer(3*nprocs)
cntk2f <- 1
for(i in 1:(3*nprocs)) {
tmp <- NULL
if (i <= nprocs) {
tmp <- model$params$probs[i]
} else if (i <= 2*nprocs) {
tmp <- model$params[["taus"]]["minus", i-nprocs]
} else {tmp <- model$params[["taus"]]["plus", i-2*nprocs]}
if(is.na(tmp)) {
K2F[i] <- cntk2f
cntk2f <- cntk2f + 1
} else if(tmp == 0 | (tmp < 1 & tmp > 0)) {
K2F[i] <- -1
} else if(tmp %in% proc_names) {
K2F[i] <- K2F[which(proc_names == as.character(tmp)) + (i>nprocs)*nprocs + (i>2*nprocs)*nprocs]
}
}
INTEGER4 <- K2F-1*(K2F!=-1)
# call C++ function RTMPT
out <- .Call("ertmpt_fit",
as.numeric(REAL),
as.numeric(REAL2),
as.numeric(REAL3),
as.character(CHAR),
as.integer(INTEGER),
as.integer(INTEGER2),
as.integer(INTEGER3),
as.integer(INTEGER4),
as.integer(INTEGER5),
as.logical(BOOL1),
as.logical(BOOL2),
as.logical(BOOL3))
# get data information
data_info <- list(Nsubj = length(unique(data_frame[,1])),
Ngroups = length(unique(data_frame[,2])),
Nprobs = sum(is.na(model$params$probs)),
Nminus = sum(is.na(model$params$taus[1,])),
Nplus = sum(is.na(model$params$taus[2,])),
Nresps = length(unique(model$responses$MAP)),
# epsilon = 1,
probs_string = names(model$params$probs[which(is.na(model$params$probs))]),
minus_string = names(model$params$probs[which(is.na(model$params$taus[1,]))]),
plus_string = names(model$params$probs[which(is.na(model$params$taus[2,]))]),
transformation = transformation)
# prepare output list
ertmpt <- list()
ertmpt$samples <- make_mcmc_list(file = out$pars_samples, infofile = infofile,
Nchains = Nchains, Nsamples = Nsamples,
data_info = data_info, keep = old_label)
out$pars_samples <- NULL; gc()
# diagnostics
ertmpt$diags <- get_diags(diag_file = diag_path, data_info = data_info, keep = old_label)
if (file.exists(diag_path)) file.remove(diag_path)
ertmpt$diags$R_hat <- gelman.diag(ertmpt$samples, multivariate = FALSE)
# specs
infos <- readinfofile(infofile)
ertmpt$specs <- list(model = model, n.chains = Nchains, n.iter = Nsamples, n.burnin = Nwarmup, n.thin = thin,
n.groups = data_info$Ngroups, n.subj = data_info$Nsubj, Irep = Irep, Rhat_max = Rhat_max,
prior_params = prior_params, infolist = infos, call = match.call(),
n.groups = max(data_frame[,2]+1))
if(exists("transformation")) {
ertmpt$specs$transformation <- transformation
}
# WAIC & LOO
if (all(out$LogLik == 0)) {
out$LogLik <- NULL
} else {
if (indices) {
suppressWarnings( temp <- get_indices_x(out$LogLik, Nchains, Nsamples, data_frame) )
ertmpt$indices <- list(WAIC = temp$WAIC, LOO = temp$LOO)
#if (save_log_lik) ertmpt$LogLik <- temp$LogLik
rm(temp)
}
if (save_log_lik) {
ertmpt$LogLik <- out$LogLik
}
}
# if (file.exists(loglik_path)) {
# if (indices) {
# suppressWarnings( temp <- get_indices_x(loglik_path, Nchains, Nsamples, data_frame) )
# ertmpt$indices <- list(WAIC = temp$WAIC, LOO = temp$LOO)
# if (save_log_lik) ertmpt$LogLik <- temp$LogLik
# rm(temp)
# } else {
# if (save_log_lik) {
# fwrite(data.table(V1 = ""), loglik_path, append = TRUE, col.names = FALSE, row.names = FALSE)
# ll_dt <- fread(file = loglik_path, skip = 0, header = FALSE)
# ertmpt$LogLik <- as.matrix(copy(ll_dt))
# }
# }
#
# }
# file.remove(loglik_path)
# remove data file
if (!keep_data_path && file.exists(data_path)) file.remove(data_path)
# remove info file and chains
if (file.exists(infofile)) file.remove(infofile)
if (file.exists(chains_path)) file.remove(chains_path)
# summary
ertmpt$summary <- writeSummaryERTMPT(ertmpt, keep = old_label)
# output
class(ertmpt) <- "ertmpt_fit"
gc()
return(ertmpt)
}
#' Fit Diffusion-RT-MPT Models
#'
#' Given model and data, this function a Hamiltonian MCMC sampler and stores the samples in an mcmc.list called \code{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 \code{diags}. Optionally, the \code{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 (\code{LogLik}) can also be stored.
#' Some specifications of the function call are also saved in \code{specs}.
#'
#' @param model A list of the class \code{drtmpt_model}.
#' @param data Optimally, a list of class \code{drtmpt_data}. Also possible is a \code{data.frame} or a
#' path to the text file. Both, \code{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 \code{\link{to_drtmpt_data}} first, which gives back a
#' \code{drtmpt_data} list with information about the changes in the data, that were needed.
#' @param n.chains Number of chains to use. Default is 4. Must be larger than 1 and smaller or equal to 16.
#' @param n.iter Number of samples per chain. Default is 1000.
#' @param n.phase1 Number of samples for phase 1 (adaptation phase). Default is 1000.
#' @param n.phase2 Number of samples for phase 2. Default is 2000.
#' @param n.thin Thinning factor. Default is 1.
#' @param Rhat_max Maximal Potential scale reduction factor: A lower threshold that needs to be reached before the actual sampling starts. Default is 1.05
#' @param Irep Every \code{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 \code{Irep}:
#' \itemize{
#' \item \code{n.phase1} is a multiple of \code{Irep}
#' \item \code{n.phase2} is a multiple of \code{Irep}
#' \item \code{n.phase1} is smaller than or equal to \code{n.phase2},
#' \item \code{Irep} is a multiple of \code{n.thin} and
#' \item \code{n.iter} is a multiple of \code{Irep / n.thin}.
#' }
#' @param prior_params Named list with prior parameters. All parameters have default values, that lead to uninformative priors. Vectors are not allowed.
#' Allowed parameters are:
#' \itemize{
#' \item \code{prec_epsilon}: prior precision for population means of process-related parameters.
#' Default is \code{1.0}.
#' \item \code{delta_df}: degrees of freedom of t-distribution for motor times. Default is \code{10}.
#' \item \code{delta_mu}: mean of t-distribution for motor times. Default is \code{0.5}.
#' \item \code{delta_scale}: scale of t-distribution for motor times. Default is \code{1.0}.
#' \item \code{SIGMA_Corr_eta}: shape parameter for LKJ distribution for process-related parameters. Default is \code{4.0}.
#' \item \code{SIGMA_SD_rho}: scale parameter of half-Cauchy distribution for process-related parameters. Default is \code{2.5}.
#' \item \code{GAMMA_Corr_eta}: shape parameter for LKJ distribution for motor-related parameters. Default is \code{4.0}.
#' \item \code{GAMMA_SD_rho}: scale parameter of half-Cauchy distribution for motor-related parameters. Default is \code{0.5}.
#' \item \code{Omega2_alpha}: shape parameter of gamma distribution for residual variance. Default is \code{0.0025}.
#' \item \code{Omega2_beta}: rate parameter of gamma distribution for residual variance. Default is \code{0.5}.
#' }
#' @param flags Either NULL or a list of
#' \itemize{
#' \item \code{old_label} If set to \code{TRUE} the old labels of "subj" and "group" of the data will be used in the elements of the output list.
#' Default is \code{FALSE}.
#' \item \code{indices} Model selection indices. If set to \code{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 \code{loo} package. If you want to have the
#' log-likelihood matrix stored in the output of this function, you can set \code{loglik} to \code{TRUE}. Default for
#' \code{indices} is \code{FALSE}.
#' \item \code{loglik} If set to \code{TRUE} and \code{indices = TRUE} the log-likelihood matrix for each iteration and trial will
#' be saved in the output as a matrix. Default is \code{FALSE}.
#' \item \code{random_init} If set to \code{TRUE} the initial values are randomly drawn. If \code{FLASE} maximum likelihood is used
#' for initial values.
#' }
#' @param control Either NULL or a list of
#' \itemize{
#' \item \code{maxthreads} for the ML estimation of the initial values and the calculation of the DIC values one can use more than
#' \code{n.chains} threads for parallelization. Default is 4 like \code{n.chians}. \code{maxthreads} must be larger or equal to
#' \code{n.chains}.
#' \item \code{maxtreedepth1_3} maxtree-depth of the no-U-turn algorithm in Phases 1 to 3
#' \item \code{maxtreedepth4} maxtree-depth of the no-U-turn algorithm in Phases 4
#' }
#' @return A list of the class \code{drtmpt_fit} containing
#' \itemize{
#' \item \code{samples}: the posterior samples as an \code{mcmc.list} object,
#' \item \code{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,
#' \item \code{specs}: some model specifications like the model, arguments of the model call, and information about the data transformation,
#' \item \code{indices} (optional): if enabled, WAIC and LOO,
#' \item \code{LogLik} (optional): if enabled, the log-likelihood matrix used for WAIC and LOO.
#' \item \code{summary} includes posterior mean and median of the main parameters.
#' }
#' @references
#' Klauer, K. C. (2010). Hierarchical multinomial processing tree models: A latent-trait approach. \emph{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.
#' \emph{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.
#' \emph{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.
#' \emph{Journal of Machine Learning Research, 11(Dec)}, 3571-3594.
#'
#' @examples
#' ####################################################################################
#' # 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
#' }
#' @author Raphael Hartmann
#' @useDynLib "rtmpt", .registration=TRUE
#' @export
#' @importFrom coda gelman.diag
#' @importFrom methods as callGeneric new is
#' @importFrom utils read.table write.table
fit_drtmpt <- function(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) {
# CHECKS
model_elmnts <- c("lines", "params", "responses")
if (!is.list(model)) stop("\"model\" must be a list.")
if (!all(model_elmnts %in% names(model))) stop("\"model\" must contain \"", model_elmnts[which(!(model_elmnts %in% names(model)))[1]], "\".")
if (n.chains < 2 || n.chains > 16) stop("\"n.chains\" must be larger than 1 and lower or equal to 16.")
if (Irep < 1) stop("\"Irep\" must be larger than or equal to one.")
if (n.thin < 1) stop("\"n.thin\" must be larger than or equal to one.")
if (n.phase1 %% Irep) stop("\"n.phase1\" must be a multiple of \"Irep\".")
if (n.phase2 %% Irep) stop("\"n.phase2\" must be a multiple of \"Irep\".")
if (n.phase1 > n.phase2) stop("\"n.phase2\" must be larger or equal to \"n.phase1\".")
if (Irep %% n.thin != 0) stop("\"Irep\" must be a multiple of \"n.thin\".")
if (n.iter %% (Irep/n.thin) != 0) stop("\"n.iter\" must be a multiple of \"Irep\" / \"n.thin\".")
if (n.iter < Irep/n.thin) stop("\"n.iter\" must be greater or equal to \"Irep\" / \"n.thin\" = ", Irep*n.thin, ".")
if (Rhat_max < 1) stop("\"Rhat_max\" must be larger than or equal to one.")
if (!is.null(prior_params) && !is.list(prior_params)) stop("\"prior_params\" must be a list or NULL.")
if (!is.null(flags) && !is.list(flags)) stop("\"flags\" must be a list or NULL.")
if (!is.null(control) && !is.list(control)) stop("\"control\" must be a list or NULL.")
if (!is.null(flags)) {
if ("old_label" %in% names(flags)) if (!is.logical(flags$old_label)) stop("\"flags$old_label\" must either be TRUE or FALSE.")
if ("indices" %in% names(flags)) if (!is.logical(flags$indices)) stop("\"flags$indices\" must either be TRUE or FALSE.")
if ("loglik" %in% names(flags)) if (!is.logical(flags$loglik)) stop("\"flags$loglik\" must either be TRUE or FALSE.")
if ("random_init" %in% names(flags)) if (!is.logical(flags$random_init)) stop("\"flags$random_init\" must either be TRUE or FALSE.")
}
if (!is.null(control)) {
if ("maxthreads" %in% control) if (!is.numeric(control$maxthreads)) stop("\"control$maxthreads\" must be numeric.")
if ("maxtreedepth1_3" %in% control) if (!is.numeric(control$maxtreedepth1_3)) stop("\"control$maxtreedepth1_3\" must be numeric.")
if ("maxtreedepth4" %in% control) if (!is.numeric(control$maxtreedepth4)) stop("\"control$maxtreedepth4\" must be numeric.")
}
# SET FLAGS
if (!is.list(flags)) flags <- list()
if (!"old_label" %in% names(flags)) {
old_label <- FALSE
} else {
old_label <- flags$old_label
}
if (!"indices" %in% names(flags)) flags$indices <- FALSE
if (!"loglik" %in% names(flags)) flags$loglik <- FALSE
if (!"random_init" %in% names(flags)) flags$random_init <- FALSE
# SET CONTROLS
if (!is.list(control)) control <- list()
if (!"maxthreads" %in% names(flags)) control$maxthreads <- 4
if (!"maxtreedepth1_3" %in% names(flags)) control$maxtreedepth1_3 <- 5
if (!"maxtreedepth4" %in% names(flags)) control$maxtreedepth4 <- 9
# PREPARE DATA
keep_data_path <- FALSE
data_frame <- NULL
if (is.data.frame(data)) {
temp_data <- to_drtmpt_data(data, model)
data_frame <- temp_data$data
if("transformation" %in% names(temp_data)) {
transformation <- temp_data$transformation
} else transformation <- list()
} else if (is.character(data)) {
temp_data <- to_drtmpt_data(read.table(file = data, header = TRUE), model)
data_frame <- temp_data$data
if("transformation" %in% names(temp_data)) {
transformation <- temp_data$transformation
} else transformation <- list()
keep_data_path <- TRUE
} else if (inherits(data, "drtmpt_data")) {
data_frame <- data$data
if("transformation" %in% names(data)) {
transformation <- data$transformation
} else transformation <- list()
}
rm(data); gc()
data_elmnts <- c("subj", "group", "tree", "cat", "rt")
if (!all(data_elmnts %in% names(data_frame))) stop("\"data\" must contain \"", data_elmnts[which(!(data_elmnts %in% names(data_frame)))[1]], "\".")
if (!all(data_elmnts == names(data_frame))) {
df <- data_frame
data_frame <- df[,match(data_elmnts, names(data_frame))]
}
if (any(is.na(data_frame)) || min(data_frame) < 0) stop("All values in \"data\" need to be larger than or equal to zero and must not be NA.")
if (max(data_frame[,1]+1) != length(unique(data_frame[,1]))) stop("\"max(data$subj+1)\" must equal \"length(unique(data$subj))\". You might want to use to_rtmpt_data().")
if (max(data_frame[,2]+1) != length(unique(data_frame[,2]))) stop("\"max(data$group+1)\" must equal \"length(unique(data$group))\". You might want to use to_rtmpt_data().")
if (max(data_frame[,3]+1) != length(unique(data_frame[,3]))) stop("\"max(data$tree+1)\" must equal \"length(unique(data$tree))\". You might want to use to_rtmpt_data().")
if (max(data_frame[,4]+1) != length(unique(data_frame[,4]))) stop("\"max(data$cat+1)\" must equal \"length(unique(data$cat))\". You might want to use to_rtmpt_data().")
# PREPARE PATHS
# Inputs
model_path <- gsub("\\\\", "/", tempfile(pattern = "model", tmpdir = tempdir(), fileext = ".info"))
mdl_path <- gsub("\\\\", "/", tempfile(pattern = "model", tmpdir = tempdir(), fileext = ".txt"))
data_path <- gsub("\\\\", "/", tempfile(pattern = "data", tmpdir = tempdir(), fileext = ".txt"))
write.table(x = data_frame, file = data_path, sep = " ", row.names = FALSE, col.names = TRUE)
# Outputs
raus_path <- gsub("\\\\", "/", tempfile(pattern = "raus", tmpdir = tempdir(), fileext = ".out"))
ll_path <- gsub("\\\\", "/", tempfile(pattern = "loglik", tmpdir = tempdir(), fileext = ".out"))
cont_path <- gsub("\\\\", "/", tempfile(pattern = "continue", tmpdir = tempdir(), fileext = ".out"))
means_path <- gsub("\\\\", "/", tempfile(pattern = "means", tmpdir = tempdir(), fileext = ".out"))
tests_path <- gsub("\\\\", "/", tempfile(pattern = "tests", tmpdir = tempdir(), fileext = ".out"))
rand_path <- gsub("\\\\", "/", tempfile(pattern = "random", tmpdir = tempdir(), fileext = ".out"))
# PREPARE INFOFILE
infofile <- try(get_infofile(model, mdl_txt = mdl_path, mdl_info = model_path))
if(is(infofile, "try-error")) stop("problem with S4 routines in makin infofile.\n")
# Prior Parameter
prior_names <- c("prec_epsilon", "delta_df", "delta_mu", "delta_scale",
"SIGMA_Corr_eta", "SIGMA_SD_rho", "GAMMA_Corr_eta", "GAMMA_SD_rho",
"Omega2_alpha", "Omega2_beta")
if (is.null(prior_params)) prior_params <- list()
if (!("prec_epsilon" %in% names(prior_params))) prior_params$prec_epsilon <- 1.0
if (!("delta_df" %in% names(prior_params))) prior_params$delta_df <- 10
if (!("delta_mu" %in% names(prior_params))) prior_params$delta_mu <- 0.5
if (!("delta_scale" %in% names(prior_params))) prior_params$delta_scale <- 1.0
if (!("SIGMA_Corr_eta" %in% names(prior_params))) prior_params$SIGMA_Corr_eta <- 4.0
if (!("SIGMA_SD_rho" %in% names(prior_params))) prior_params$SIGMA_SD_rho <- 2.5
if (!("GAMMA_Corr_eta" %in% names(prior_params))) prior_params$GAMMA_Corr_eta <- 4.0
if (!("GAMMA_SD_rho" %in% names(prior_params))) prior_params$GAMMA_SD_rho <- 0.5
if (!("Omega2_alpha" %in% names(prior_params))) prior_params$Omega2_alpha <- 0.0025
if (!("Omega2_beta" %in% names(prior_params))) prior_params$Omega2_beta <- 0.5
if (!is.numeric(unlist(prior_params))) stop("prior_params must be a list of numerics and integers")
if (prior_params$delta_df %% 1 != 0) stop("delta_df must be an integer")
# PROCESS NAMES AND NUMBER
proc_names <- names(model$params$threshold)
nprocs <- length(proc_names)
# KERN2FREE
K2F <- integer(3*nprocs)
cntk2f <- integer(3)
consts <- rep(NaN, 3*nprocs)
comp <- rep(FALSE, 3*nprocs)
cntcomp <- integer(3)
for (i in 1:3) {
cntk2f[i] <- 0
for (j in 1:nprocs) {
tmp <- NULL
if (i == 1) {
tmp <- model$params$threshold[1, j]
} else if (i == 2) {
tmp <- model$params$driftrate[1, j]
} else if (i == 3) {
tmp <- model$params$startpoint[1, j]
}
ind <- (i-1)*nprocs+j
if(is.na(tmp) | is.numeric(tmp)) {
K2F[ind] <- cntk2f[i]
cntk2f[i] <- cntk2f[i] + 1
if (is.na(tmp)) {
comp[(cntk2f[i]-1)*3 + i] <- TRUE
}
if (is.numeric(tmp)) {
consts[(cntk2f[i]-1)*3 + i] <- tmp
}
} else if(tmp %in% proc_names) {
K2F[ind] <- K2F[which(proc_names == as.character(tmp)) + (i>=2)*nprocs + (i==3)*nprocs]
}
}
}
# PATHS ARGUMENT
CHAR1 <- as.character(c(Data = data_path,
Model = infofile,
Raus = raus_path,
LogLik = ll_path,
Continue = cont_path,
Means = means_path,
Tests = tests_path,
Random = rand_path,
tmpdir = paste0(tempdir(), "/")))
# MCMC SETTINGS ARGUMENT
INTEGER1 <- as.integer(c(IREP = Irep,
PHASE1 = n.phase1,
PHASE2 = n.phase2,
THIN = n.thin,
NOTHREADS = n.chains,
SAMPLE_SIZE = n.chains*n.iter,
MAXTHREADS = control$maxthreads,
nKERN = dim(model$responses)[1],
nPROCS = nprocs,
respno = length(unique(model$responses$MAP)),
cat2resp = model$responses$MAP))
REAL1 <- c(Rhat_max = Rhat_max)
# FLAGS ARGUMENT
BOOL1 <- as.integer(c(DIC_FL = flags$indices,
LOG_LIK_FL = (flags$loglik),
ML_INIT_FL = !flags$random_init))
# PRIORS ARGUMENT
INTEGER2 <- as.integer(c(degf = prior_params$delta_df))
REAL2 <- c(PRIOR = prior_params$prec_epsilon,
etat = prior_params$SIGMA_Corr_eta,
taut = prior_params$SIGMA_SD_rho,
etar = prior_params$GAMMA_Corr_eta,
taur = prior_params$GAMMA_SD_rho,
mu_prior = prior_params$delta_mu,
rsd = prior_params$delta_scale,
prioralpha = prior_params$Omega2_alpha,
priorbeta = prior_params$Omega2_beta)
# HMC OPTIONS ARGUMENT
INTEGER3 <- as.integer(c(maxtreedepth1_3 = control$maxtreedepth1_3,
maxtreedepth4 = control$maxtreedepth4))
# CONTINUATION ARGUMENT
INTEGER4 <- as.integer(c(goon = FALSE,
ADDITION = 0))
# CONSTANTS AND EQUALIZATION
INTEGER5 <- as.integer(c(kern2free = K2F,
comp = comp,
ifree = cntk2f))
REAL3 <- c(consts = consts)
# C++ FUNCTION DRTMPT CALL
out <- .Call("drtmpt_fit", CHAR1, INTEGER1, REAL1, BOOL1,
INTEGER2, REAL2,
INTEGER3, INTEGER4,
REAL3, INTEGER5)
file.remove(data_path)
# DATA INFORMATION
data_info <- list(Nsubj = length(unique(data_frame[,1])),
Ngroups = length(unique(data_frame[,2])),
Nthreshold = sum(is.na(model$params$threshold[1,])),
Ndriftrate = sum(is.na(model$params$driftrate[1,])),
Nstartpoint = sum(is.na(model$params$startpoint[1,])),
Nresps = length(unique(model$responses$MAP)),
thresh_string = names(model$params$threshold[which(is.na(model$params$threshold[1,]))]),
drift_string = names(model$params$driftrate[which(is.na(model$params$driftrate[1,]))]),
start_string = names(model$params$startpoint[which(is.na(model$params$startpoint[1,]))]),
transformation = transformation)
# PREPARE OUTPUT LIST
drtmpt <- list()
drtmpt$samples <- make_mcmc_list_d(file = out$pars_samples, infofile = infofile,
Nchains = n.chains, Nsamples = n.iter,
data_info = data_info, keep = old_label)
file.remove(raus_path)
# DIAGNOSTICS
drtmpt$diags <- get_diags_d(diag_file = tests_path, data_info = data_info, keep = old_label, DIC = flags$indices)
drtmpt$diags$R_hat <- gelman.diag(drtmpt$samples, multivariate = FALSE)
# SPECS
infos <- readinfofile(infofile)
drtmpt$specs <- list(model = model, n.chains = n.chains, n.iter = n.iter, n.phase1 = n.phase1, n.phase2 = n.phase2,
n.thin = n.thin, n.groups = data_info$Ngroups, n.subj = data_info$Nsubj, Irep = Irep,
Rhat_max = Rhat_max, prior_params = prior_params, infolist = infos, call = match.call())
if(exists("transformation")) {
drtmpt$specs$transformation <- transformation
}
# # COMPUTE WAIC AND LOO-IC
# if (flags$indices) {
# if (file.exists(ll_path)) {
# if (flags$indices) {
# temp <- NULL
# suppressWarnings( temp <- get_indices_x(log_lik = ll_path, Nchains = n.chains, Nsample = n.iter, df = data_frame) )
# drtmpt$indices <- list(WAIC = temp$WAIC, LOO = temp$LOO)
# if (flags$loglik) drtmpt$LogLik <- temp$LogLik
# rm(temp)
# } else {
# if (flags$loglik) {
# fwrite(data.table(V1 = ""), ll_path, append = TRUE, col.names = FALSE, row.names = FALSE)
# ll_dt <- fread(file = ll_path, skip = 0, header = FALSE)
# drtmpt$LogLik <- as.matrix(copy(ll_dt))
# }
# }
# file.remove(ll_path)
# }
# }
# WAIC & LOO
if (all(out$loglik == 0)) {
out$loglik <- NULL
} else {
if (flags$indices) {
suppressWarnings( temp <- get_indices_x(out$loglik, n.chains, n.iter, data_frame) )
drtmpt$indices <- list(WAIC = temp$WAIC, LOO = temp$LOO)
rm(temp)
}
if (flags$loglik) {
drtmpt$loglik <- out$loglik
}
}
# CONTINUATION
if (FALSE) {
drtmpt$continue <- readLines(con = cont_path)
}
file.remove(cont_path)
# SUMMARY
drtmpt$summary <- writeSummaryDRTMPT(x = drtmpt, keep = old_label)
# CLEAN-UP
file.remove(means_path)
file.remove(tests_path)
file.remove(rand_path)
file.remove(infofile)
file.remove(mdl_path)
# OUTPUT
class(drtmpt) <- "drtmpt_fit"
gc()
return(drtmpt)
}
Try the rtmpt package in your browser
Any scripts or data that you put into this service are public.
R Package Documentation
Browse R Packages
We want your feedback!
Note that we can't provide technical support on individual packages. You should contact the package authors for that.
Embedding an R snippet on your website
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.