Initialize all parameters for the variational inference

Initialize all parameters for the variational inference

initVEM(cliqueList, sigma_O, MO, r)

Arguments

cliqueList

List of size r of initial neighbors for each missing actor.
sigma_O

PLNmodels output: estimated observed bloc of the variance-covariance matrix.
MO

PLNmodels output: estimated variational mean values of the latent parameters corresponding to observed species.
r

Number of missing actors.

Value

Computes the following initial matrices:

  • Wginit: variational edges weights matrix.

  • Winit: edges weights matrix.

  • omegainit: precision matrix.

  • MHinit: variational mean values for the hidden latent Gaussian parameters.

Examples

data=generate_missing_data(n=100,p=10,r=1,type="scale-free", plot=TRUE)

PLNfit<-norm_PLN(data$Y)
MO<-PLNfit$MO
sigma_O=PLNfit$sigma_O
#-- find initial clique using your favorite method
findclique=FitSparsePCA(MO,r=1)
initClique=findclique$cliques
#-- initialize the VEM
initList=initVEM(cliqueList=initClique,sigma_O=sigma_O, MO=MO,r=1 )
str(initList)

#> List of 4
#>  $ Wginit   : num [1:11, 1:11] 0 0.00826 0.00826 0.00826 0.00826 ...
#>  $ Winit    : num [1:11, 1:11] 0 0.00826 0.00826 0.00826 0.00826 ...
#>  $ omegainit: num [1:11, 1:11] 5.886 0.65 0.773 0.271 -0.209 ...
#>  $ MHinit   : num [1:100, 1] -0.0113 -0.1041 0.4011 -0.57 -0.6013 ...