Releases: diyabc/abcranger
Releases · diyabc/abcranger
Massively improved oob weights computation (parallelization)
OOB weights computations in parameter estimation wasn't multhreaded. They are now.
last minute fix : invalid reading of mutationnel parameters in diyabc reftable.
Polishing param estim output stats, PLS reverting to full compute
- Global/Local inlined with abcrf (for nmae in particular)
- cosmetics
- Parallelization of forest quantiles computing for oob samples
- New
--plsmaxvar
(default 0.90) option in estimparam, see the README
Fixed a bug with oob weights computation (EstimParam)
1.14.1 Bug with oob weights in estimparam fixed
EstimParam output overhaul + PLS heuristic tuning
Fixed global values in EstimParam
1.13.1 Fixed wrong global in EstimParam
Groups added for model choice
- option
-g
for groups as asked in #38 - changed
--ntest
option name for--noob
- composite parameter documented
Updates, minor bugs
- updated to latest compiler (clang 9.0, apple 11.0, msvc 16.4)
- updated latest dependancies and vcpkg's packages (range v3 v0.9 breaking changes)
- solved some crashes, especially on macos
Various fixes, probably publication ready
Long list of changes, in short :
- Various under the hood optimisations and cmake refactoring.
- Various fixes to match exactly the output of
abcrf
R package - PLS stopping heuristic fixed
- Merge of two binaries in one
abcranger
- automated KS tests for two methodologies
- estimparam : added "local mae"
Parameter Estimation: fixed quantiles and variance
Pls selection overhaul, build system almost "user-friendly"
In this release :
- Parameter estimation precision display with 6 digits (fix #8)
- PLS new heuristic selection as discussed there #9
Plus :
- reworked the whole vcpkg dependancy and now it's bootstraped to be downloaded and launched directly in the cmake build
- made the MKL and static stuff (which currently requires clang and libc++ on linux) optional
Hopefully anybody could compile the project with a c++17 compliant compiler. - static binary with threaded MKL provided (linux only at the moment) : blazing fast PLS!