fix for ML example

[mcreel]

The ML example in the documentation (https://julianlsolvers.github.io/Optim.jl/stable/#examples/generated/maxlikenlm/) has some problems. The t-statistics are not correct, as can be verified by computing the OLS estimates. The error is that the transformation from log sigma to sigma is not correctly done. The Hessian matrix is computed after directly exponentiating the log sigma parameter, while the likelihood function is not prepared to accept sigma rather than log sigma, so the resulting Hessian is not correct. This PR uses the delta method to do the transformation.

[pkofod]

Thanks, Looks good. But ForwardDiff appear to not be in the manifest.

[mcreel]

Yes, I saw the error. Where should I add it, at the top level of Optim.jl?

[codecov]

Codecov Report

Merging #956 (57460cd) into master (24d5941) will increase coverage by 0.03%.
The diff coverage is n/a.

@@            Coverage Diff             @@
##           master     #956      +/-   ##
==========================================
+ Coverage   83.52%   83.55%   +0.03%     
==========================================
  Files          42       42              
  Lines        3022     3022              
==========================================
+ Hits         2524     2525       +1     
+ Misses        498      497       -1     

Impacted Files	Coverage Δ
...ultivariate/solvers/zeroth_order/particle_swarm.jl	99.17% <0.00%> (+0.41%)	⬆️

---

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 24d5941...57460cd. Read the comment docs.

[pkofod]

Great, thanks! I'll fix the nightly errors shortly.