- August 30, 2022
- update R and variancePartition dependency versions
- August 22, 2022
- remove dependency on PRIMME package
July 26, 2021
- fix bug in calculating effective degrees of freedom:
- report m = mean edf per response
- for fixed effect, now report
m instead of m + 1
- Add more unit tests
- in
pcTransform() varFrac is 1 but default
- enable
pca by default, and compute it only once in mvForwardStepwise()
Major changes
July 13, 2021
- add
fastApprox argument that is exact for fixed effects and approximate if random effects are included
- include
pcTransform() function
- this can be 100x faster for large datasets
- better estimate of lambda and logML
- use
lambda = 0.01 by default
- defaults are new
criterion = "BIC", shrink.method = "EB", nparamsMethod = "edf"
- remove scaling of responses to allow use of
pcTransform() on responses
- Using effect degrees of freedom is faster since it is built into
dream()
- fixed bugs handling edf for each regression
- remove redundant calculations
- Add more unit tests
October 26, 2020 - enforce compatability with variancePartition >= 1.19.20
May 5, 2020 - add back gdf
May 1, 2020 - Add bestSubsetSearch() - Add estimateMVN_EB for Empirical Bayes estimation from Multivariate Normal Inverse Wishart model (MNIWEB)
May 1, 2020 - add rough code for empirical Bayes covariance estimation
April 27th, 2020 - Add other selection criteria
April 26th, 2020 - Add fast shrinkage of eigen values for Touloumis_equal