ccas: Infer Communication Content and Structure
In matthewjdenny/CCAS: Communication Content and Network Structure

Description Usage Arguments Value Examples

Performs inference on the content conditional structure of a text valued communication network.

ccas(formula, interaction_patterns = 4, topics = 40, alpha = 1,
  beta = 0.01, iterations = 1000, metropolis_hastings_iterations = 500,
  final_metropolis_hastings_burnin = 50000,
  final_metropolis_hastings_iterations = 1e+05, thin = 1/100,
  target_accept_rate = 0.25, tolerance = 0.05,
  LSM_proposal_variance = 0.5, LSM_prior_variance = 1, LSM_prior_mean = 0,
  iterations_before_t_i_p_updates = 5, update_t_i_p_every_x_iterations = 5,
  adaptive_metropolis = TRUE, adaptive_metropolis_update_size = 0.05,
  seed = 12345, adaptive_metropolis_every_x_iterations = 1000,
  stop_adaptive_metropolis_after_x_updates = 50,
  slice_sample_alpha_m = FALSE, slice_sample_step_size = 1,
  parallel = FALSE, cores = 2, output_directory = NULL,
  output_name_stem = NULL, generate_plots = TRUE, verbose = TRUE)

`formula`	A formula object of the form 'ComNet ~ euclidean(d = 2)' where d is the number of dimensions in the latent space that the user would like to include, and ComNet is an object of class 'ComNet' generated by the prepare_data() function. This object will contain all of the relevant information about the corpus so that multiple models may be specified without the need to repreprocess the data. The formula may also include optional terms 'sender("covariate_name")', 'receiver("covariate_name")', 'nodemix("covariate_name", base = value)' and 'netcov("network_covariate")', which are defined analogously to the arguments in the latentnet package.
`interaction_patterns`	The number of different interaction patterns governing message sending and recieving under the model. Defaults to 4.
`topics`	The number of topics to be used in the model. Defaults to 40.
`alpha`	The hyperparameter governing document-topic distributions. Lower values encourage more peaked distributions. Defaults to 1.
`beta`	The hyperparameter governing the Dirichlet prior on the topic-word distributions. Lower values encourage more peaked distributions. Defaults to 0.01.
`iterations`	The number of iterations of Metropolis-within-Gibbs sampling to be used in model estimation. Defaults to 1,000.
`metropolis_hastings_iterations`	The number of Metropolis Hastings iterations to be run during each iteration of Metropolis-within-Gibbs sampling to update interaction pattern parameters. Defaults to 500.
`final_metropolis_hastings_burnin`	The number of iterations of Metropolis Hastings to run after completing the main iterations of Gibbs sampling to discard before keeping samples. Defaults to 50,000.
`final_metropolis_hastings_iterations`	The number of iterations to run Metropolis Hastings after completing all main iterations of Gibbs sampling. Defaults to 100,000. This additional number of iterations is required to ensure that the Markov chain of interaction pattern parameters has mixed appropriately and converged to the target distribution. This can occasionally take on the order of 1-10 million iterations, but is often much faster in practice.
`thin`	The proportion of network samples to keep from the final run of Metropolis Hastings to convergence. Defaults to 1/100, meaning that every 100'th network sample will be returned.
`target_accept_rate`	The target acceptance rate for the Metropolis Hastings algorithm. Defaults to 0.25 which is standard in the literature.
`tolerance`	The tolerance for differences between the observed and target Metropolis Hastings accept rates (+-). Defaults to 0.05.
`LSM_proposal_variance`	The Metropolis Hastings proposal variance for all interaction pattern parameters. Defaults to .5.
`LSM_prior_variance`	The variance of the multivariate normal prior on all interaction pattern parameters. Defaults to 1.
`LSM_prior_mean`	The mean of the multivariate normal prior on all interaction pattern parameters. Defaults to 0.
`iterations_before_t_i_p_updates`	The number of iterations to wait before beginning updates to topic interaction pattern assignments. Defaults to 5. If the user does not wish to update these assignments, the value can be set to a value greater than 'iterations'.
`update_t_i_p_every_x_iterations`	The number of iterations between updates to topic interaction pattern assignments. Defaults to 5.
`adaptive_metropolis`	Logical indicating whether adaptive Metropolis should be used (whether the proposal variance should be optimized). Defaults to TRUE.
`adaptive_metropolis_update_size`	The amount by which the Metroplis Hastings proposal distribution variance is changed (up or down) durring adaptive Metropolis. Defaults to 0.05.
`seed`	The seed to be used (for replicability across runs). Defaults to 12345.
`adaptive_metropolis_every_x_iterations`	The nubmer of iterations between proposal variance updates during the final run of Metropolis Hastings to convergence. Defaults to 1000.
`stop_adaptive_metropolis_after_x_updates`	The number of Metropolis Hastings proposal variance updates to complete during the final run of Metropolis Hastings to convergence before fixing its value. Defualts to 50. Make sure that the selection of this number is such that the proposal variance is fixed after burnin.
`slice_sample_alpha_m`	Logical indicating whether hyperparameter optimization should be used to determine the optimal value of alpha. Defaults to FALSE. If TRUE, then alpha_m will be slice sampled. This can improve model fit.
`slice_sample_step_size`	The initial size of the slice to use when slice sampling alpha (hyperparameter optimization). Defaults to 1.
`parallel`	Argument indicating whether the token topic distributions should be generated in parallel. Defaults to FALSE. Can significantly reduce runtime when training a model with a large number of topics.
`cores`	The number of cores to be used if the parallel option is set to TRUE. Should not exceed the number of cores available on the machine and will not show performance gains if cores > topics.
`output_directory`	The directory where the user would like to store output from the model. Defaults to NULL. If NULL, then the current working directory will be used to store output if an output_name_stem is provided.
`output_name_stem`	Defaults to NULL. If not NULL, then output will be saved to disk using the output_name_stem to differentiate it from output from other model runs.
`generate_plots`	Logical indicating whether diagnostic and summary plots should be generated, defaults to TRUE.
`verbose`	Defaults to TRUE, if FALSE, then no output is printed to the screen by the inference code.

An object of class CCAS containing estimation results.

## Not run: 
set.seed(12345)
# read in data prepared by the prepare_data() function.
data(ComNet_data)
# specify a formula that we will use for testing.
formula <- ComNet_data ~ euclidean(d = 2) +
           nodemix("Gender", base = "Male")
CCAS_Object <- ccas(formula,
                    interaction_patterns = 4,
                    topics = 40,
                    alpha = 1,
                    beta = 0.01,
                    iterations = 20,
                    metropolis_hastings_iterations = 500,
                    final_metropolis_hastings_iterations = 10000,
                    final_metropolis_hastings_burnin = 5000,
                    thin = 1/10,
                    target_accept_rate = 0.25,
                    tolerance = 0.05,
                    adaptive_metropolis_update_size = 0.05,
                    LSM_proposal_variance = .5,
                    LSM_prior_variance = 1,
                    LSM_prior_mean = 0,
                    slice_sample_alpha_m = TRUE,
                    slice_sample_step_size = 1,
                    generate_plots = TRUE,
                    output_directory = NULL,
                    output_name_stem = NULL)

## End(Not run)