goldfish.latent

goldfish.latent extends the goldfish package with models that include latent variables. Therefore, how to define the event sequence data objects for modeling and the effects available for modeling uses goldfish’s data objects definitions and statistics effects. Inferences of the proposed models are made using Hamiltonian Chain Monte Carlo as implemented in Stan.

Installation

You can install the development version of goldfish.latent like so:

remotes::install_github("snlab-ch/goldfish.latent", build_vignettes = TRUE)

Example

The first model introduced in goldfish.latent is the DyNAM with random effects. Model formulation allows to include monadic statistic or actors’ covariates to explain the variability of the random effects.

library(goldfish.latent)

# using cmdstanr for getting posterior samples and 
library(cmdstanr)

# using goldfish social evolution data set
library(goldfish)
data("Social_Evolution")
callNetwork <- defineNetwork(nodes = actors, directed = TRUE) |>
 linkEvents(changeEvent = calls, nodes = actors)
callsDependent <- defineDependentEvents(
 events = calls, nodes = actors, defaultNetwork = callNetwork
)
data2stan <- CreateData(
 randomEffects = list(inertia ~ 1),
 fixedEffects = callsDependent ~ recip + trans
)

stanCode <- CreateModelCode(data2stan)

mod01 <- cmdstan_model(stanCode)
mod01Samples <- mod01$sample(
  data = data2stan[["dataStan"]],
  parallel_chains = 4, chains = 4,  iter_warmup = 500, iter_sampling = 500,
  show_messages = FALSE
)

Using cmdstanr functionalities makes easier fit to use summary and plotting function from packages that works with MCMC posterior samples. For the summary of the posterior samples cmdstanr uses the posterior package. Plots from the posterior samples can be done using, for example, bayesplot package.

mod01Samples$summary("beta")
#> # A tibble: 3 × 10
#>   variable   mean median    sd   mad     q5    q95  rhat ess_bulk ess_tail
#>   <chr>     <dbl>  <dbl> <dbl> <dbl>  <dbl>  <dbl> <dbl>    <dbl>    <dbl>
#> 1 beta[1]   4.94   4.92  0.424 0.394  4.30  5.67   1.00      793.     936.
#> 2 beta[2]   1.63   1.63  0.207 0.202  1.30  1.98   1.00     2881.    1372.
#> 3 beta[3]  -0.272 -0.268 0.211 0.204 -0.640 0.0649 0.999    3188.    1797.

  # names fixed effects coincides with colnames(data2stan$dataStan$X)

The ComputeLogLikelihood() function allows computing the marginal or conditional log-likelihood for the model using MCMC samples from the posterior distribution. Parallel computation using parallel package is possible. Using the loo functionalities is possible to compute the Watanabe Information Criterion waic() and the Leave-One-Out approximation using Pareto smoothed importance sampling loo(). Those information criteria can be used to compare models using loo_compare().

logLikMod01 <- ComputeLogLikelihood(mod01Samples, data2stan, spec = 4)

# loo and waic computation using the marginal
library(loo)

relEff <- relative_eff(exp(logLikMod01))
looM01 <- loo(logLikMod01, r_eff = relEff)
#> Warning: Some Pareto k diagnostic values are too high. See help('pareto-k-diagnostic') for details.
looM01
#> 
#> Computed from 2000 by 34 log-likelihood matrix
#> 
#>          Estimate    SE
#> elpd_loo   -679.8 112.0
#> p_loo        17.7   8.0
#> looic      1359.7 224.1
#> ------
#> Monte Carlo SE of elpd_loo is NA.
#> 
#> Pareto k diagnostic values:
#>                          Count Pct.    Min. n_eff
#> (-Inf, 0.5]   (good)     28    82.4%   231       
#>  (0.5, 0.7]   (ok)        0     0.0%   <NA>      
#>    (0.7, 1]   (bad)       3     8.8%   8         
#>    (1, Inf)   (very bad)  3     8.8%   1770      
#> See help('pareto-k-diagnostic') for details.

Code of Conduct

Please note that the goldfish.latent project is released with a Contributor Code of Conduct. By contributing to this project, you agree to abide by its terms.