- 1.22 (3/10/2012) Two fixes:
- Stability fix for the block relaxation algorithm, which
was found to output negative parameter values for a large K
value. Thanks to Matthew for finding this.
- Fst estimates were upward biased; have now switched to the
method of Reynolds et al. (1983). Thanks to Suyash for
- 1.21 (7/10/2011): Fix a bug in supervised analysis (thanks to
Dienekes for finding this) and a bug in dealing with large K
(≥20) (found by Mait and Jessica). For those who are interested,
here is a description of the latter bug.
- Note to users: the matrix formerly
referred to as F, containing the population allele frequencies,
is now known (as of v1.20) as P, and is output to a .P file.
This is to bring our notation in line with that of the STRUCTURE
- 1.20 (6/21/2011): Several improvements:
- Cross-validation is now k-fold and uses binomial deviance
residuals to estimate prediction error;
- now compiled as a 64-bit binary allowing bigger data files,
and giving a ~20% speedup
- minor bug fixes.
- 1.12 (4/23/2011): Fix memory leak in bootstrap code
- 1.11 (3/17/2011): Statically link some libs for Mac OS X, for
people without the dev tools installed
- 1.1 (2/8/2011): Parallel processing, supervised analysis. Minor speedups and cleanups.
- 1.04 (6/17/2010): Added cross-validation for choosing K. Also a 10% speedup from improved cache locality in the core algorithm.
- 1.03 (4/15/2010): Floating point improvements. Now is about twice as fast as original release 1.00 (used in paper)
- 1.02 (8/9/2009): Streamlined quadratic programming code, reducing runtime ~30%; reduced memory consumption
- 1.01 (8/7/2009): Added support for PLINK binary pedigree (.bed) files
- 1.00 (7/1/2009): Initial public release