Various Ways of fitting a 'GBLUP' model using BGLR
In the following example we show how to fit a GBLUP model (i.e., a Gaussian process) using different parameterization.
* [Using oringial inputs (e.g., SNPs)](#BRR)
* [Using a G-matrix (or kernel)](#RKHS)
* [Using eigenvalues and eigenvectors](#RKHS2)
* [Using scaled-principal components)](#PC)
* [Using a Cholesky decomposition](#CHOL)
* [Using a QR decomposition](#QR)
---------------------------------------------
**(i) Providing the markers, using `model='BRR'`**
In this case BGLR asigns iid normal priors to the marker effects.
wzxhzdk:0
[Menu](#menu)
---------------------------------------------
**(2) Providing the G-matrix**
BGLR Fits these Gaussian models using the eigenvalue decomposition og G. The eigenvalue decomposition is computed internally using
`eigen()`.
wzxhzdk:1
[Menu](#menu)
---------------------------------------------
**(3) Providing eigenvalues and eigenvectors**
This strategy can be used to avoid computing the eigen-decomposition internally. This can be useful if a model will be fitted several times (e.g., cross-validation).
wzxhzdk:2
[Menu](#menu)
---------------------------------------------
**(4) Providing scaled-eigenvectors and using `model='BRR'`**
wzxhzdk:3
[Menu](#menu)
---------------------------------------------
**(5) Using the Cholesky decompositon and `model='BRR'`**
This approach won't work if G is not positive definite; in our case the matrix is positive semi-definite, we can make it positive definite by adding a small constant to the diagonal.
wzxhzdk:4
[Menu](#menu)