[Bug?] Use Stick breaking or other scheme instead of division by sum to avoid division by small or zero value

[SteveBronder]

For the code here

    if(maxSR | maxSTARR | CSMratio) cvxr_wts <- cvxr_wts / sum(cvxr_wts)
    cvxr_wts <- as.vector(cvxr_wts)
    cvxr_wts <- normalize_weights(cvxr_wts)

If the optimization ends up with weights that sum to near zero or zero the line that divides by the sum could cause NA or very large values. Would it be alright if I made a PR to add something like a stick breaking scheme (example here) that would avoid the division?

[braverock]

@SteveBronder normally I would recommend optimizing long portfolios and short portfolios hierarchically. This is more likely to result in robust solutions.

That would reduce but doesn't eliminate the chances of the problem you describe here.

Yes, a PR and tests would be appreciated to demonstrate the problem and solution.

[SteveBronder]

This is more likely to result in robust solutions.

By robust do you mean "more likely to be a numerically valid solution"? I always get confused by that word.

I'll try to make an example this weekend

[SteveBronder]

Oh one more question about this line

if(maxSR | maxSTARR | CSMratio) cvxr_wts <- cvxr_wts / sum(cvxr_wts)

If someone has leverage set to not one, this ignores that and would essentially make leverage 1 right? Should this only be run if a user has a leverage of 1?