ELBO calculation in the presence of missing data

[gaow]

This is only problematic for EZ model alpha=1 not for EE model alpha=0. Here is the data-set and notebook that reproduces the problem. In the notebook I've also done some initial diagnosis.

issue_6.tar.gz

Notice that here Y has missing data. But I doubt it is problem with missing data because the residual variance I use here is diagonal.

[gaow]

But I doubt it is problem with missing data because the residual variance I use here is diagonal.

Another place missing data will have an impact is the quantity XtX and occurs precisely only in the code that computes ELBO:

https://github.com/gaow/mmbr/blob/master/R/susie_regression.R#L129

So ELBO computation in the presence of missing data should be rederived.

[gaow]

For the time being I added a warning message here:

https://github.com/gaow/mmbr/blob/f2a2cf3df719dd1725b13b61525006e7b8bf20b1/R/ibss_algorithm.R#L129