closes #129 by fixing the interface after #123

[tlienart]

PR #123 makes an explicit distinction between two cases:

1. binary classification with an explicit LogisticClassifier
2. binary classification with an explicit MultinomialClassifier

The way these things are processed is not the same. In short, the output of LC is a score corresponding to "class 1" (whatever it is), the score comes out of sigmoid (logic), whereas in the MN case it comes out of a softmax.

Anyway long story short, the PR #123 was not properly finished (the MLJ.fit was fixed correctly but the MLJ.predict should have been fixed in basically the same way.

[codecov-commenter]

Codecov Report

Merging #130 (50f5572) into dev (5bb7c6d) will decrease coverage by 1.42%.
The diff coverage is 66.66%.

@@            Coverage Diff             @@
##              dev     #130      +/-   ##
==========================================
- Coverage   96.07%   94.64%   -1.43%     
==========================================
  Files          21       20       -1     
  Lines         866      878      +12     
==========================================
- Hits          832      831       -1     
- Misses         34       47      +13     

Impacted Files	Coverage Δ
src/mlj/interface.jl	81.03% <66.66%> (-7.68%)	⬇️
src/mlj/classifiers.jl	50.00% <0.00%> (-50.00%)	⬇️
src/mlj/regressors.jl	6.66% <0.00%> (-49.59%)	⬇️
src/fit/iwls.jl	100.00% <0.00%> (ø)
src/glr/d_l2loss.jl	100.00% <0.00%> (ø)
src/glr/d_robust.jl	100.00% <0.00%> (ø)
src/glr/d_logistic.jl	100.00% <0.00%> (ø)
src/loss-penalty/standard.jl	100.00% <0.00%> (ø)
src/MLJLinearModels.jl
... and 3 more

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