make mean(::Chi) more robust

[bicycle1885]

Current:

julia> mean(Chi(1000))
NaN

PR:

julia> mean(Chi(1000))
31.614871896968094

[codecov-commenter]

Codecov Report

Merging #1305 (db05bd3) into master (3512557) will increase coverage by 0.00%.
The diff coverage is 100.00%.

@@           Coverage Diff           @@
##           master    #1305   +/-   ##
=======================================
  Coverage   81.55%   81.55%           
=======================================
  Files         115      115           
  Lines        6641     6642    +1     
=======================================
+ Hits         5416     5417    +1     
  Misses       1225     1225           

Impacted Files	Coverage Δ
src/univariate/continuous/chi.jl	95.34% <100.00%> (ø)
src/univariate/discrete/geometric.jl	80.00% <0.00%> (+0.28%)	⬆️

---

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 3512557...db05bd3. Read the comment docs.

[devmotion]

Thanks, this seems reasonable 👍 Can you add the MWE as a test?

[bicycle1885]

I tried to generate the continuous_test.ref.json file on my local environment, but it failed with the following error:

...
On Pareto(3.0, 2.0) --> Pareto: alpha=3 beta=2 
On Rayleigh() --> Rayleigh: sigma=1 
On Rayleigh(3.0) --> Rayleigh: sigma=3 
On Rayleigh(8.0) --> Rayleigh: sigma=8 
On StudentizedRange(2.0, 2.0) --> StudentizedRange: nu=2 k=2 
Error in distr$pdf(xs) : attempt to apply non-function
Calls: do.main -> eval.points
3: (function () 
   {
       traceback(2)
       q()
   })()
2: eval.points(distr)
1: do.main("continuous_test")

I'm using R 4.0.3. Perhaps newer than expected? Sorry, I'm not so familiar with the R language.

[devmotion]

Oh, I didn't realize that it's mainly/only tested with the R/json setup. I thought you could just add a check that the example above does not return NaN anymore, possibly in a new Julia test file (although, of course, it could also be added to the R-based checks).

[bicycle1885]

Great. That's much easier. I've added a test in a new file chi.jl.

[devmotion]

Thank you @bicycle1885!