powder: Power posterior sampling using differential evolution
In jeff324/powder: Power posterior sampling via differential evolution

Description Usage Arguments Value Examples

powder Runs power posterior sampling using differential evolution markov chain monte carlo

powder(model, data, num.temps = 30, alpha = 0.3, high.temps.first = FALSE,
  n.sequences = 1, current.sequence = 1, n.samples = 1000,
  n.chains = NULL, burnin = 500, meltin = 250, de_params = list(),
  method = "standard", return.samples = TRUE, verbose = TRUE,
  update = 100)

`model`	See `LBA` for more information and an example.
`data`	A list of data where each element is the data for a specific subject
`num.temps`	Number of temperatures (i.e. number of power posteriors to sample from)
`alpha`	controls the temperature schedule. 0.3 is recommended.
`high.temps.first`	If true, then the posterior will be sampled from first, followed by lower temperature posteriors. If false, then the prior will be sampled from first, followed by higher temperature posteriors.
`n.sequences`	This is useful for running parallel power posteriors. If `n.sequences = 1`, then all power posteriors will be run sequentially. If `n.sequences > 1` then `num.temps / n.sequences` power posteriors will be run sequentially.
`current.sequence`	If n.sequences > 1, then `num.temps / n.sequences` power posteriors will be run sequentially. `current.sequence` is the index of the current sequence. For example, if `num.temps = 10`, `n.sequences = 2`, and `current.sequence = 1` then temperatures 1 through 5 will be run. If `current.sequence = 2` then temperatures 6 through 10 will be run.
`n.samples`	The number of samples to draw from each power posterior
`n.chains`	The number of chains to use. By default, `n.chains` is the number of subject-level parameters times 3.
`burnin`	The number of samples to discard when computing the marginal likelihood. These samples are included in the raw output.
`meltin`	The power posteriors are sampled sequentially. When moving to the next power posterior, the sampling process takes some time to adapt to the new power posterior. `meltin` is the number of samples to discard after moving to the next power posterior. While `burnin` discards samples from the first power posterior, `meltin` discard samples for all other power posteriors.
`de_params`	A list containing the following options for DE-MCMC. See Turner et al. (2013) for details. `b` The parameter for the uniformly distributed noise term for the DE proposal. Default is .001. `migration` A `logical`. Indicates whether to circulate the states of the chains to remedy problem of outlier chains. Default is FALSE. `migration.freq` A `numeric`. Number of iterations to wait between each migration step. `migration.start` A `numeric`. Iteration to start migrating. This should be after chains are burned in. `migration.end` A `numeric`. Iteration to stop migrating. Migration should stop well before sampling is finished. `randomize_phi` A `logical`. Assume independence between subject and group-level parameters. Useful for improved sampling of group-level parameters. Default is FALSE. `zLag` A `numeric`. When `randomize_phi = TRUE` and `method = 'parallel'`, zLag is the number of iterations that can be reached into from the past for the z-Update. `zStart` A `numeric`. When `randomize_phi = TRUE` and `method = 'parallel'`, zStart is the iteration to begin z-Updating.
`method`	A character vector that specifies the type of sampling to be performed and accepts one of the following: `standard` This option samples from each power posterior along the temperature schedule in sequence `parallel` This option samples from each power posterior in parallel, where the target density of each chain is a power posterior at a given temperature. Using this method will cause n.chains to default to num.temps. `posterior` This option samples from the posterior (i.e. the power posterior where temperature = 1). Although this option is useful for parameter estimation, it is not possible to obtain marginal likelihoods via this option. `wbic` This option option samples from a single power posterior, where the temperature = 1/log(n) (Watanabe, 2013).
`return.samples`	If true, return subject and group-level samples. Otherwise, return only the log likelihoods from each power posterior.
`verbose`	Display progress
`update`	Number of iterations between progress display updates

A list with the following elements.

log.like.list A list containing the log likelihoods for each temperature

theta An array containing the subject-level samples

phi An array containing the group-level samples

Note, if return.samples = FALSE then only log.like.list will be returned.

## Not run: 
model = LBA.Individual$new(b=T)
data('individual',package='powder')
pow.out = powder(data=individual, model=model, num.temps=30)
est = marginal.likelihood(pow.out)
print(est)

## End(Not run)