Skip to contents

Hyperparameter and MCMC control settings for spbgfm

Source: R/spbgfm_control.R
spbgfm_control.Rd

Collects the fixed hyperparameters of Sp-BGFM together with the Metropolis-Hastings tuning constants into a single list. All values have defaults matching those used in the simulation studies of Zhang, Shen, Chen and Lee; override individual entries by passing them as named arguments.

Usage

spbgfm_control(
  a_phi = NULL,
  a_tau = 0.1,
  b_tau = 1,
  a_sig = 3,
  b_sig = 3,
  u2_beta = 3,
  Lr = 30,
  L_alpha = 35,
  a_psi_r = 1,
  a_psi_alpha = 1,
  a_w = 5,
  b_w = 5,
  a_w_alpha = 5,
  b_w_alpha = 5,
  ur2 = 1,
  u2_xi_r = 1,
  w_bound = 9.9e-17,
  acc_tar = 0.234
)

Arguments

a_phi, a_tau, b_tau

Dirichlet-Horseshoe hyperparameters for the factor loadings. a_phi is the Dirichlet concentration; when NULL (the default) it is set from the data to 1/(0.2 * J), the rule given in the paper, which is about 1/20 for \(J \approx 100\). tau_k ~ Gamma(a_tau, b_tau / J) with defaults a_tau = 0.1 and b_tau = 1.

a_sig, b_sig

Inverse-gamma prior parameters for the domain-specific idiosyncratic variance \(v_m^2\). Default 3 and 3.

u2_beta

Prior variance of the regression coefficients \(\beta_{mjp}\). Default 3. Ignored when no covariate is supplied.

Lr, L_alpha

Stick-breaking truncation levels for the sample size factor \(r_{im}\) and the baseline abundance \(\alpha_{smj}\). Defaults 30 and 35.

a_psi_r, a_psi_alpha

Dirichlet process total mass parameters \(c^r\) and \(c^\alpha\). Default 1.

a_w, b_w, a_w_alpha, b_w_alpha

Beta prior parameters for the inner mixture weights \(\omega^r_{ml}\) and \(\omega^\alpha_{ml}\). Default 5 each.

ur2

Kernel variance \(u_r^2\) of the size-factor mixture. Default 1.

u2_xi_r

Prior variance \(u^2_{\xi^r}\) of the size-factor atoms \(\xi^r_{ml}\). Default 1.

w_bound

Numeric in (0, 0.5). The inner mixture weights are confined to [w_bound, 1 - w_bound]. The mean-constrained atom \((\nu - \omega\xi)/(1 - \omega)\) diverges as \(\omega \to 1\), so a bound keeps the mixture atoms in a numerically sensible range. Default 9.9e-17, reproducing the scripts that accompany the paper; a value such as 1e-3 is more conservative.

acc_tar

Target acceptance rate for the adaptive Metropolis-Hastings update of \(\phi_k\). Default 0.234.

Value

A named list of class "spbgfm_control".

References

Zhang, S., Shen, Y., Chen, I. A. and Lee, J. (2025). Sparse Bayesian Group Factor Model for Feature Interactions in Multiple Count Tables Data. Journal of the American Statistical Association, 120(550), 723–736. doi:10.1080/01621459.2025.2449721

Examples

ctrl <- spbgfm_control(a_psi_r = 3, a_psi_alpha = 3)
str(ctrl)
#> List of 18
#>  $ a_phi      : NULL
#>  $ a_tau      : num 0.1
#>  $ b_tau      : num 1
#>  $ a_sig      : num 3
#>  $ b_sig      : num 3
#>  $ u2_beta    : num 3
#>  $ Lr         : num 30
#>  $ L_alpha    : num 35
#>  $ a_psi_r    : num 3
#>  $ a_psi_alpha: num 3
#>  $ a_w        : num 5
#>  $ b_w        : num 5
#>  $ a_w_alpha  : num 5
#>  $ b_w_alpha  : num 5
#>  $ ur2        : num 1
#>  $ u2_xi_r    : num 1
#>  $ w_bound    : num 9.9e-17
#>  $ acc_tar    : num 0.234
#>  - attr(*, "class")= chr "spbgfm_control"