Closes #1213 added logpdf for BernoulliLogit

[HarrisonWilde]

…Logit to distributions.jl

[devmotion]

I'm not sure if separate types are necessary here, wouldn't just another constructor be sufficient?

I also don't think one should define separate vector-valued distributions since one can define it easily as a product of univariate distributions (in fact, I think VecBinomialLogit should be removed).

[HarrisonWilde]

That is fair enough, agreed on the vector distributions, I did think it unnecessary but thought I would follow the lead of BinomialLogit, should I remove all of them then?

And what do you mean by another constructor sorry?

[codecov]

Codecov Report

Merging #1214 into master will decrease coverage by 0.05%.
The diff coverage is 0.00%.

@@            Coverage Diff             @@
##           master    #1214      +/-   ##
==========================================
- Coverage   65.64%   65.59%   -0.06%     
==========================================
  Files          25       25              
  Lines        1278     1279       +1     
==========================================
  Hits          839      839              
- Misses        439      440       +1     

Impacted Files	Coverage Δ
src/Turing.jl	100.00% <ø> (ø)
src/stdlib/distributions.jl	52.38% <0.00%> (-1.28%)	⬇️

---

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 d5c47f1...7b96e50. Read the comment docs.

[devmotion]

That is fair enough, agreed on the vector distributions, I did think it unnecessary but thought I would follow the lead of BinomialLogit, should I remove all of them then?

Just don't add VecBernoulliLogit, I'd say 😃

And what do you mean by another constructor sorry?

Sorry, I should have explained it better. I meant that maybe it would be sufficient to add the line

BernoulliLogit(logitp::Real) = BinomialLogit(1, logitp)

and export BernoulliLogit. Then users would be able to use BernoulliLogit but actually internally we would still use the BinomialLogit distribution, so it would just be an additional convenient constructor for BinomialLogit.