add covariance_estimator

[matthieugomez]

Return the CovarianceEstimator used to compute the variance-covariance matrix

[nalimilan]

Maybe we should just add a CovarianceEstimator method? Not sure. Cc: @kleinschmidt

[kleinschmidt]

Does statsbase own CovarianceEstimator type?

[nalimilan]

Yes:

StatsBase.jl/src/cov.jl

Line 158 in 24a80ca

CovarianceEstimator

[kleinschmidt]

A few people on slack suggested that the constructor pun would be more appropriate when some kind of computation is involved, which (as far as I can tell) isn't the intended use case here, right? This would just retrieve an existing <:CovarianceEstimator. So I'm inclined to think that covariance_estimator seems more natural.

[matthieugomez]

I'm personally indifferent, so let me know.

[nalimilan]

Hard to decide... Another question if we decide to add a function: maybe it should be vcov_estimator since it goes with vcov?

[codecov]

Codecov Report

Merging #530 into master will decrease coverage by 0.04%.
The diff coverage is 0%.

@@            Coverage Diff             @@
##           master     #530      +/-   ##
==========================================
- Coverage    90.5%   90.45%   -0.05%     
==========================================
  Files          21       21              
  Lines        2095     2096       +1     
==========================================
  Hits         1896     1896              
- Misses        199      200       +1

Impacted Files	Coverage Δ
src/StatsBase.jl	100% <ø> (ø)	⬆️
src/statmodels.jl	49.54% <0%> (-0.46%)	⬇️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update b039107...9e1a66f. Read the comment docs.

[matthieugomez]

Or use covariance instead of vcov ;)

[nalimilan]

Yeah, well, that would be breaking, so... An even more appealing solution would be to overload cov from Base, since that's really the same thing. Then you'd call the function covestimator, which is a bit shorter.

[matthieugomez]

I don't think it's that important actually. Maybe we should close until we have a better story for covariance vcov etc?

[nalimilan]

You mean you don't really need that function? Then yes, maybe better close until we have a use case, that's always better to design good APIs.

[matthieugomez]

Exactly