If I set build_vignettes=T, the package fails to install. Error message:

Error in loadVignetteBuilder(pkgdir, TRUE) :
vignette builder 'R.rsp' not found
Execution halted

When applying group updating algorithm to fit bcoxph models, an offset will be induced as an element of the model object. The offset element is a by-product of the internal call of coxph by treating non-updating coefficients and its linear predictors as an offset term. However, without replacing the artificial offset with the original offset. It creates a problem when acquiring data within cv.bh when cross-validating the fitted model, see

Hence, it will produce an error message when calling cv.bh

Error in cv.bh.coxph(object = object, nfolds = nfolds, foldid = foldid, : 'data' not given in object

Minimum reproducible Example:
The following code is a reproducible example where group-updated bcoxph causes the problem while jointly updated does not.

library(BhGLM)
library(survival)

## Example Code from the BhGLM Manual
N = 1000
K = 100
x = sim.x(n=N, m=K, corr=0.6) # simulate correlated continuous variables  
h = rep(0.1, 4) # assign four non-zero main effects to have the assumed heritabilty 
nz = as.integer(seq(5, K, by=K/length(h))); nz
#> [1]  5 30 55 80
yy = sim.y(x=x[, nz], mu=0, herit=h, p.neg=0.5) # simulate responses
yy$coefs
#>         x5        x30        x55        x80 
#>  0.4094061 -0.4152127  0.4186130 -0.3971476

y = yy$y.surv
d = table(y[,2]); d[1]/sum(d) # cencoring proportion
#>     0 
#> 0.486

ps = 0.05
mdl_joint = bcoxph(y ~ ., data = x, prior = De(0, ps))
mdl_group = bcoxph(y ~ ., data = x, prior = De(0, ps), method.coef = 50)

joint_cv <- cv.bh(mdl_joint)  # No error
#> Fitting ncv*nfolds = 10 models: 
#> 1 2 3 4 5 6 7 8 9 10 
#>  Cross-validation time: 0.215 minutes
group_cv <- cv.bh(mdl_group)  # Error caught
#> Error in cv.bh.coxph(object = object, nfolds = nfolds, foldid = foldid, : 'data' not given in object

^{Created on 2022-01-03 by the reprex package (v2.0.1)}

Thanks for the great package!

I find it sometimes useful to set a known 0 intercept, especially when in applications the model involves some categorical predictors and one wants to use dummy coding other than the sort of default reference coding. Can we open the intercept argument in glmnet to the user in bmlasso?

In the case of dummy coding, we might also don't want to penalize some of the predictors that will be interpreted as group means. Is it possible to have something like the no.penalty that will set penalty.factor being 0 in the glmnet part of the algorithm (might also change the updating rule of theta)?

Thanks!
Yunyi

Hey guys, congratulations for your paper. I would like to rerun your analysis with a different pathway database. Could you please point me to the scripts to rerun the analysis?
Thank you in advance,

Regards

Daniel