Head over to the Issues page and ask a question, suggest a new feature or report a bug.
If your data is particularly complex and you currently have it saved as a list then the following conversion should move it into a pmwg compatible format.
In this first case you start with a list of subjects, and each subject has one or more data objects associated with it.
# Play data - a list of data.frames associated with subject names in this case - but each list item should be able to be any object
data_list <- list(subject1 = data.frame(group = c('high', 'low'), meanrt = c(0.8, 1.4)), subject2 = data.frame(group = c('high', 'low'), meanrt = c(0.9, 1.1)))
# Turn that list into a tibble, which preserves the internal data.frame objects
data <- tibble(subject = names(data_list), df = data_list)
# Alternatively, use the I() function to interpret the list elements as is
data <- data.frame(subject = names(data_list), df = I(data_list))
# Then pass the data.frame/tibble as documented to the pmwgs object, and run your sampling stages.
# The data passed into your log likelihood function would have the following form: (basically the output from either of the last two links above)
# # A tibble: 1 x 2
# subject df
# <chr> <named list>
# 1 subject1 <df[,2] [2 × 2]>
# So you will also need to extract the internal data.frame or other object in your log likelihood function, using something like:
df <- data$df[[data$subject]]
# Then you can work with the data.frame (or whever format your list elements are) from there within the rest of your log-likelihood function.Alternatively if you have a list of studies, and each study has a subject column you can follow the following format to make your dataset compatible with pmwg
# Play data - a list of data.frames associated with task names in this case.
data_list <- list(task1 = data.frame(subject = c('A', 'B'), meanrt = c(0.8, 1.4)), task2 = data.frame(subject = c('A', 'B'), meanrt = c(0.9, 1.1)))
# We'll use this helper function to split a larger tibble by subject
split_tibble <- function(x) {
lapply(x, function(x){
split(x, x$task)
})
}
# Now we can rearrange the data_list above to be in the correct format. We first bind our list to a new data.frame with a task column, then split it by subject to get our top level subject column.
# Then we turn this list from the split operation into a tibble and turn each list in the df column to be a named list with the names representing the different tasks
data <- data_list %>%
bind_rows(.id = "task") %>%
split(.$subject) %>%
tibble(subject = names(.), df = .) %>%
mutate(df = split_tibble(df))
# The data passed into your log likelihood function would have the following form: (basically the output from either of the last two links above)
# # A tibble: 1 x 2
# subject df
# <chr> <named list>
# 1 A <named list [2]>
# So you will also need to extract the internal named list or other object in your log likelihood function, using something like:
data <- data$df[[data$subject]]
# From this point you can calculate the likelihood for task1 using the data.frame data$task1, and similarly for task2 and then sum the log-likelihoods to get the joint likelihood value to return to pmwgIf you are getting low acceptance rates (as displayed in the progress bar) sometimes this could be caused by one or two participant datasets.
The internal function accept_rate will return an array of mean acceptance rates, where the acceptance rate is the number of new, unique, particles accepted by the sampler
The default is to calculate this mean over the last 200 samples (or as many samples as exist if fewer than 200). The following two calls show how to do this for a custom window, or for all samples.
# For a sampled object called sampler
pmwg:::accept_rate(sampler, window_size=500)
# For all samples
pmwg:::accept_rate(sampler, window_size=sampler$samples$idx)