This page gathers some small programs or routines, with limited documentation.
If you like it, cite it!
Simple program to generate B-splines functions for given range or given values of covariable, with choice of equi-distant splines or user-defined knots.
Fortran 90 rewrite of subroutine DKMWHF (derived from Henderson's DJNVHF) to obtain a generalised inverse of a real, symmetric matrix.
Simple front-end program to DKMXHF which reads in the matrix from a file & writes out the inverse to another file (DKMXHF not included).
Complete program for downloading (ginverse.f90 with dkmxhf.f90) ginverse.zip
Hash tables are a standard computational tool to facilitate fast `look-up' in a table or structured list.
A common task in analyzing quantitative genetic data is to identify the unique levels of an effect (with many different levels), e.g. animal IDs, in large data sets. Hash tables provide an effective way to carry out this task, and recode levels in running order (1 to N) when no numerical ordering of the original codes is required.
Here is a simple FORTRAN 95 module (derived from original code by Bruce Tier) which provides all the necessary routines to use one or more hash lists for such task:
And here is a small example program illustrating how to use it:
Download both: hash.zip
An approximate inverse of a sparse (symmetric and positive definite) matrix – or selected elements thereof - can be obtained using Gibbs sampling, as described by Harville (1999).
Here is a FORTRAN 95 module which contains the necessary subroutines to store the non-zero elements of the sparse matrix in linked list format, sort it and obtain the approximate inverse. The version given determines the diagonal elements of the inverse only, but is readily extended to return diagonal blocks or a complete inverse. It uses routines from the
ranlibf package (Texas Medical Center) to sample random normal deviates (not included; other RNG software can readily be substituted):
And here is a small test program illustrating its use:
Download both: harville.zip
For symmetric, positive definite matrices stored in compressed format (see George and Liu, 1981) & assuming that matrices have been re-ordered to reduce fill-in and the symbolic factorisation has been carried out.
Computations require BLAS and LAPACK libraries. Fortran subroutines: