CTWMCMCBRIDGE Build CTW tree graph(s) and make MCMC samples.
Y = CTWMCMCBRIDGE(X,OPTS) returns both the analytic (or weighted)
entropy estimation on the context-tree weighted (CTW) graph(s), and
numerous Markov chain Monte Carlo (MCMC) Bayesian samples of the CTW
tree graph(s), which are derived from the input data in X. The type of
analytic entropy calculated depends on the options, as does the number
of MCMC samples to return. X is a single element cell array, like that
obtained with DIRECTBIN, whose rows represent stimulus repeats and
whose columns represent time bins, that is a matrix of binned spike
trains. The output Y is a structure containing both analytic and MCMC
entropy estimates. This function bridges the functions CTWMCMCTREE and
CTWMCMCSAMPLE, without outputing the CTW tree graph(s).
The options and parameters for this function are:
OPTS.beta: Krischevsky-Trofimov ballast parameter used in the
calculation of local codelength, Le, which also serves as the
Dirichlet prior parameter in subsequent Markov chain Monte Carlo
(MCMC) tree sampling. Its value should be greater than 0. The
default is 1/A, where A is the largest value in the input data X
plus one.
OPTS.gamma: The weighting between tree node and its children, used
when calculating the weighted codelength, Lw. Its value should
lie between 0 and 1, non-inclusive. The default is 0.5.
OPTS.max_tree_depth: The maximum tree depth (may be used to conserve
memory). Its value must be greater than 0. The default is 100000.
OPTS.h_zero: Flag to indicate use of the H_zero estimator for
deterministic nodes, that is such nodes will not be weighted when
true. Note that this value is not used in tree building, per say,
but in subsequent MCMC sampling. The default is 1 (true).
OPTS.tree_format: The default and only allowable value is 'none'. To
output CTW tree graphs, please see CTWMCMCTREE.
OPTS.memory_expansion: The ratio by which tree memory is expanded
when reallocation become necessary during tree building. Its
value must be greater than or equal to 1. The default is 1.61.
OPTS.nmc: The number of MCMC samples to make. Its value must be
greater than 0, and should be at least 100. The default is 199.
OPTS.entropy_estimation_method: A cell array of entropy estimation
methods. Please see the Spike Train Analysis Toolkit
documentation for more information, and corresponding entropy
options. The default is {'plugin'}.
OPTS.variance_estimation_method: A cell array of variance
estimation methods. Please see the Spike Train Analysis Toolkit
documentation for more information, and corresponding variance
options (listed with entropy options). The default is not to
perform any variance estimation.
OPTS.mcmc_iterations: The absolute number of iterations to run the
Markov chain Monte Carlo simulation (for each OPTS.nmc sample).
If OPTS.mcmc_min_acceptances probability vectors have been
accepted, this is also the minimum number of iterations. The
default is 100.
OPTS.mcmc_max_iterations: The maximum number of Markov chain Monte
Carlo iterations. The simulation runs OPTS.mcmc_iterations sized
batches of iterations until OPTS.mcmc_min_acceptances probability
vectors are accepted, or this number is reached. The default is
10000.
OPTS.mcmc_min_acceptances: The minimum number of Markov chain Monte
Carlo acceptances, that is the number of acceptable probability
vectors. The default is 20.
Y = CTWMCMCBRIDGE(X) uses the default options and parameters.
[Y,OPTS_USED] = CTWMCMCBRIDGE(X) or [Y,OPTS_USED] =
CTWMCMCBRIDGE(X,OPTS) additionally return the options used.
See also DIRECTBIN, CTWMCMC, CTWMCMCTREE, CTWMCMCSAMPLE, CTWMCMCINFO.
This function calls: