Adding missing features of cosine distribution

[jaksle]

This is a very straightforward PR. It adds missing mgf, cgf, cf and direct affine transformations for the Cosine distribution. There are simple formulas for these, so there is no reason not to include them.

One thing which might require attention is if I protect from dividing by 0 correctly.

[codecov-commenter]

Codecov Report

❌ Patch coverage is 73.68421% with 5 lines in your changes missing coverage. Please review.
✅ Project coverage is 86.34%. Comparing base (e5d4fef) to head (6c2e5f9).

Files with missing lines	Patch %	Lines
src/univariate/continuous/cosine.jl	73.68%	5 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #2017      +/-   ##
==========================================
- Coverage   86.36%   86.34%   -0.03%     
==========================================
  Files         146      146              
  Lines        8788     8807      +19     
==========================================
+ Hits         7590     7604      +14     
- Misses       1198     1203       +5     

☔ 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]

Can you add tests for the new functionality?

[jaksle]

I cleaned cf too. I also noticed the parameters are used once now, so their unpacking can be made even simpler.

I added testing against some known simple values. Some more are needed or is this enough?

My code was based on how the rest of cosine.jl looked like. When I saw your corrections the code above started looking suspicious for me and indeed, pdf, cdf and ccdf were type unstable. So I corrected them, though I'm not sure if my solution is optimal.

I'm also not sure about dependencies, but its' easy to change in any case.

[jaksle]

@devmotion Could I ask for checking if now it is okay? I am a little unsure about the tests. Is there an instruction somewhere what is the recommended way to write them in Distributions.jl?

[devmotion]

In general, it is unnecessary and usually it is incorrect to precompute any return types.

For testing, IIRC some utilities such as test_mgf or test_affine (?) exists.

[jaksle]

@devmotion Good to know that such type inference is not useful. I removed the changes related to this.

Optionally, maybe something like

function pdf(d::Cosine{T}, x::Real) where T<:Real
    z = (x - d.μ) / d.σ
    y = (1 + cospi(z)) / (2d.σ)
    return insupport(d, x) ? y : zero(y)
end

could be considered.

I also added application of build-in functions to test affine transformations and cgf.