Fits DBN with low-rank sender effects using batch FFBS
Usage
dbn_lowrank_accurate(
Y,
family = c("ordinal", "gaussian", "binary"),
r = 2,
nscan = 10000,
burn = 1000,
odens = 1,
ar1_alpha = TRUE,
update_rho_alpha = FALSE,
ar1_B = FALSE,
update_rho_B = FALSE,
seed = 6886,
verbose = TRUE,
time_thin = 1,
previous = NULL,
init = NULL,
symmetric = FALSE,
...
)Arguments
- Y
Data array (nodes x nodes x relations x time)
- family
Data family
- r
Rank for low-rank factorization. A good starting point is
ceiling(log2(n))wherenis the number of nodes. Increase if posterior predictive checks show poor fit. Default: 2.- nscan
Number of iterations of the Markov chain (beyond burn-in)
- burn
Burn-in for the Markov chain
- odens
Output density for the Markov chain
- ar1_alpha
Use AR(1) for alpha dynamics
- update_rho_alpha
Update AR coefficient for alpha
- ar1_B
Use AR(1) for B dynamics
- update_rho_B
Update AR coefficient for B
- seed
Random seed
- verbose
Logical or numeric. TRUE prints every 100 iterations, numeric prints every n iterations, FALSE suppresses output.
- time_thin
Save every nth time point
- previous
Previous results to continue from
- init
Initial values
- symmetric
Logical. Not supported for low-rank models (will error). Default: FALSE.
- ...
Additional arguments (currently unused)
See also
dbn for the main dispatcher, param_summary for posterior summaries
Examples
# \donttest{
sim <- simulate_lowrank_dbn(n = 8, time = 5, r = 2, seed = 1)
fit <- dbn_lowrank(sim$Y, r = 2, nscan = 200, burn = 100, verbose = FALSE)
# }