TODO

[mmkim1210]

• Rename package to MRVCModels.jl and register package
• Implement Fisher scoring algorithms, which can be faster and get more accurate estimates in some cases
• Make inference by information matrix faster (e.g. avoid or reduce storages_nd_nd alloc if possible)
• Handle $\boldsymbol{\Omega} = \sum_{i=1}^{m} \boldsymbol{\Gamma}_i * \boldsymbol{V}_i$, where $*$ is Khatri-Rao product of $d \times d$ partition matrices. Also handle different sets of covariates. Statistical model in this paper is subsumed under this case.
• Implement missing response model for MRTVCModel → not possible
• Implement EM algorithm for MRTVCModel
• Eigenvalue decomposition for single variance component case
• Simplify computation for V[end] = I
• Implement simulation function
• GLMM
• Low rank updates
• Make output look more aesthetically pleasing (e.g. parse.jl)