Integrating ZIPLN to PLNmodels

[jchiquet]

Zero inflation with V-EM by mean field approximation.

What is done

• Zero inflation with 3 parametrizations
  • "single" : common, global ZI parameter
  • "row": row-wise ZI parameter (vector with n entries)
  • "col": column-wise ZI parameter (vector with p entries)
  • "covar" : covariate dependent ZI component
• ZI with 4 covariance models
  • "full" : fully parametrized residual covariance
  • "diagonal" : diagonal residual covariance
  • "spherical" : spherical residual covariance
  • "sparse": sparse inverse covariance ($\ell_1$ based penalty)

In progress

• proper inclusion and testing of covariates in ZI
• finish first batch of testing
• double check the documentation
• Check and add all required R6 and S3 methods for all ZI objects (predict, coef, sigma)

Latter

• Use close form for estimation W|Y by implementing LambertW function
• Network-like methods for ZIPLNfit_sparse
• ZIPLNnetwork, a collection with PLNfamilly collecting a series of ZIPLNfit_sparse object with automatic grid