Code accompanying Stochastic Lanczos estimation of genomic variance components for linear mixed-effects models.
-
L_Seed
constructs bases for Krylov subspaces: B, (A+σI)B, (A+σI)²B, ... -
L_Solve
solves (A + σI)X = B using results fromL_Seed
-
SLQ_LDet
returns approximate log det (A + σI) given spectal decompositions of Jacobi matrices from Lanczos decompositions of seed Krylov subspaces for probe vectors -
L_FOMC_REML
extension of BOLT-LMM algorithm to recycle Krylov subspace bases involved in solving linear systems -
SLDF_REML
zero order REML estimation using (shifted) block Lanczos conjugate gradients and (shifted) stochastic Lanczos quadrature
High performance variant of Stochastic First-order Derivative-Free REML algorithm (SLDF_REML
) as employed in Assortative mating biases marker-based heritability estimators.
-
L_Seed_numba
compiled variant ofL_Seed
above -
L_Solve_numba
compiled variant ofL_Seed
above -
SLQ_LDet
see above -
SLDF_REML_numba
compiled variant ofSLDF_REML
above
In contrast to SLDF_REML
, SLDF_REML_numba
assumes all covariates have been projected out of the genotype and phenotype data. See the SL_REML_v0.1b
directory for installation instructions and a toy example.