Differences
This shows you the differences between two versions of the page.
| Both sides previous revision Previous revision | |||
|
wombat:specialanalyses [2018/08/30] kmeyer ["Single step" genomic BLUP] |
wombat:specialanalyses [2018/08/30] (current) kmeyer ["Single step" genomic BLUP] |
||
|---|---|---|---|
| Line 51: | Line 51: | ||
| - The first is simply an analysis in which the inverse of the combined relationship matrix, ${\bf H}^{-1} $, is suplied in a ''*.gin'' file. It is invoked with the run option ''----s1step''. \\ | - The first is simply an analysis in which the inverse of the combined relationship matrix, ${\bf H}^{-1} $, is suplied in a ''*.gin'' file. It is invoked with the run option ''----s1step''. \\ | ||
| * As for run option ''– –solvit'', the mixed model equations are set up once and stored in core. | * As for run option ''– –solvit'', the mixed model equations are set up once and stored in core. | ||
| - | * A special feature for the single step analysis is that the part of the coefficient matrix in the mixed model equations pertaining to genotyped animals is stored as a dense submatrix. This implies substantial RAM requirements for large analyses, but allows efficient, multi-threaded linrary routine for dense matrix manipulations to be exploited. | + | * A special feature for the single step analysis is that the part of the coefficient matrix in the mixed model equations pertaining to genotyped animals is stored as a dense submatrix. This implies substantial RAM requirements for large analyses, but allows efficient, multi-threaded library routine for dense matrix manipulations to be exploited. |
| * Iterative solutions are obtained using a pre-conditioned conjugate gradient algorithm with a choice of diagonal, block-diagonal of SSOR (default) preconditioning scheme. | * Iterative solutions are obtained using a pre-conditioned conjugate gradient algorithm with a choice of diagonal, block-diagonal of SSOR (default) preconditioning scheme. | ||
| - A low RAM alternative is invoked via run option ''----s2step'': This employs a PCG algorithm with diagonal preconditioner, using 'iteration on data' instead of in-core storage of the mixed model equations. \\ It requires pedigree information (to set up the inverse of the numerator relationship matrix) and the 'add-on' part in the combined relationship matrix, $ {\bf G}^{-1} - {\bf A}_{22}^{-1}$ as a ''*.gin'' file. | - A low RAM alternative is invoked via run option ''----s2step'': This employs a PCG algorithm with diagonal preconditioner, using 'iteration on data' instead of in-core storage of the mixed model equations. \\ It requires pedigree information (to set up the inverse of the numerator relationship matrix) and the 'add-on' part in the combined relationship matrix, $ {\bf G}^{-1} - {\bf A}_{22}^{-1}$ as a ''*.gin'' file. | ||
