Multivariate Bayesian variable selection regression

Check susie coverage

Here I change coverage from default 95% to various other numbers for susie CS and check the proportion of "false" CS.

The computation is coded in the coverage workflow of this notebook.

In [1]:
%revisions -s
Revision Author Date Message
431e3c8 Gao Wang 2018-06-07 Add comparison table
507ab0e Gao Wang 2018-06-06 Add coverage check
In [2]:
%cd ~/GIT/github/mvarbvs/dsc

/home/gaow/GIT/github/mvarbvs/dsc
In [3]:
print_summary = function(dat) {
    res = c(0,0)
    for (level in c(0.01, 0.05, 0.1, 0.15, 0.2, 0.25)) {
        s = as.character(level*100)
        d = do.call(rbind, lapply(1:length(dat), function(i) dat[[as.character(i)]][[s]]))
        d = colSums(d)
        res = rbind(res, c(level, d[2]/(d[1]+d[2])))
    }
    res = data.frame(res[-1,])
    colnames(res) = c ('1 - coverage', 'false discovery proportion')
    res
}

susie var(Y)

In [4]:
dat = readRDS('susie_comparison/Coverage_0623_estvar_false.rds')
print_summary(dat)
1 - coveragefalse discovery proportion
0.01 0.01971831
0.05 0.03490401
0.10 0.06068376
0.15 0.09037162
0.20 0.11083333
0.25 0.14107884

susie est_var

In [5]:
dat = readRDS('susie_comparison/Coverage_0623_estvar_true.rds')
print_summary(dat)
1 - coveragefalse discovery proportion
0.01 0.02604167
0.05 0.06033058
0.10 0.09813463
0.15 0.12529928
0.20 0.16891356
0.25 0.20689655
Copyright © 2016-2020 Gao Wang et al at Stephens Lab, University of Chicago