Swap out Optim.jl NelderMead.jl

[jwscook]

This PR removes Optim.jl as a dependency and replaces it with NelderMead.jl given the context of https://github.com/jwscook/NelderMead.jl/issues/7.

A point worth noting is that the size of the simplex is taken to be the standard deviation of the starting position (or 1 in the case that the standard deviation is zero). There may be a better simplex size given the range of sensible values of $\alpha_0$; please advise.

Closes https://github.com/jwscook/NelderMead.jl/issues/7.

[codecov-commenter]

Codecov Report

Merging #155 (1655db2) into master (222d30b) will increase coverage by 0.22%. The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master     #155      +/-   ##
==========================================
+ Coverage   95.16%   95.38%   +0.22%     
==========================================
  Files          27       27              
  Lines         889      889              
==========================================
+ Hits          846      848       +2     
+ Misses         43       41       -2     

Impacted Files	Coverage Δ
src/transforms/projectionpursuit.jl	96.80% <100.00%> (ø)
src/transforms/sample.jl	100.00% <0.00%> (+5.55%)	:arrow_up:

Help us with your feedback. Take ten seconds to tell us how you rate us. Have a feature suggestion? Share it here.

[jwscook]

Thank you @jwscook for the PR ❤️ Regarding the initial guess, it is guaranteed to have norm 1. Do you have suggestions for the simplex size in this case?

I'm not entirely sure what the right size should be, but after some trial and error a value of 1/2 seems to work.

[juliohm]

That is lovely @jwscook. I will just wait for the tests to pass once again and will then merge it. Did you consider moving NelderMead.jl to JuliaOptim or some other organization with multiple maintainers? It should get more visibility there. :)