Implement std(::MultivariateDistribution)

[bicycle1885]

This implements std for multivariate distributions as an element-wise square root of variance: std(d) = sqrt.(var(d)).

[codecov-commenter]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 86.27%. Comparing base (8bb4181) to head (363bf57).

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #1352      +/-   ##
==========================================
+ Coverage   86.25%   86.27%   +0.01%     
==========================================
  Files         146      146              
  Lines        8766     8777      +11     
==========================================
+ Hits         7561     7572      +11     
  Misses       1205     1205              

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[devmotion]

Seems reasonable! Can you also add std to the documentation of multivariate distributions?

[st--]

@bicycle1885 are you still interested in finishing this PR? @devmotion made a reasonable code suggestion that you would just need to accept, the other thing that seems missing is "add std to the documentation of multivariate distributions" - @devmotion could you clarify what you mean by that ?

[devmotion]

Unfortunately, it seems in general var(d) might not be mutable - eg it can be a Fill array (eg for Dirichlet) or it could be a static array. So either

• we live with these sometimes unnecessary allocations in the general fallback until Julia provides a native way to distinguish mutable and immutable arrays (similar to ArrayInterface) and demand that one has to implement a more efficient alternative for specific distributions if possible, or
• we demand that var returns a mutable array in general and one has to implement std for distributions for which var is immutable.

Personally, I think it might be better to have a more efficient default implementation and eg. add a fix for Dirichlet.

[devmotion]

@mschauer or @andreasnoack do you think it's fine to add a special case for Dirichlet or should it assume immutability by default? I suggested map! to reduce allocations (#1352 (comment)) but this is problematic for Dirichlet due since Fill arrays are not mutable.

[devmotion]

@bicycle1885 are you still interested in finishing this PR? @devmotion made a reasonable code suggestion that you would just need to accept, the other thing that seems missing is "add std to the documentation of multivariate distributions" - @devmotion could you clarify what you mean by that ?

I addressed both. I also added a fix for Dirichlet, let's see if tests pass.

[st--]

@devmotion tests are passing, should it get merged now ?

[devmotion]

I don't want to approve and merge my own changes. This does not feel right.

[azev77]

@devmotion @st-- in open source ppl write PRs then forget about them, moving on to other things ALL THE TIME!
If we wait for people, we'll never get anything done