JuliaML / LossFunctions.jl

Julia package of loss functions for machine learning.

https://juliaml.github.io/LossFunctions.jl/stable

Other

147 stars 33 forks source link

Deprecate bang! variants towards a cleaner API #133

Closed juliohm closed 4 years ago

juliohm commented 4 years ago

This PR addresses an initial step towards a cleaner API as discussed in #126. It contains a simple change that already shows major benefits in the test suite, and that will certainly help us to advance more quickly with our plans of improvement.

@joshday I appreciate if you can review this work. I'd also like to share the next steps that I am planning to work on in the following weeks after this PR is reviewed and merged:

Reconsider the special case with SparseArrays for MarginLoss that lives in sparse.jl
Refactor the test suite to improve coverage and to facilitate future contributions
Replace the AggMode types by simple Julia functions for greater flexibility
Update the docstrings of the currently available losses, and use plot recipes in the docs
Consider UnicodePlots.jl in printing.jl given that it is a very lightweight dependency

Looking forward to keep working on it.

Best,

codecov-commenter commented 4 years ago

Codecov Report

Merging #133 into master will decrease coverage by 0.13%. The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master     #133      +/-   ##
==========================================
- Coverage   93.88%   93.75%   -0.14%     
==========================================
  Files          11       11              
  Lines         572      560      -12     
==========================================
- Hits          537      525      -12     
  Misses         35       35

Impacted Files	Coverage Δ
src/LossFunctions.jl	`100.00% <ø> (ø)`
src/sparse.jl	`100.00% <100.00%> (ø)`
src/supervised.jl	`100.00% <100.00%> (ø)`

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 1de4efd...ec8e199. Read the comment docs.

joshday commented 4 years ago

Deprecated methods still need to be exported/tests should stay. My time for reviewing JuliaML PRs is much more limited going forward.

juliohm commented 4 years ago

Thank you @joshday for your time and contributions. I will take the lead moving forward with the maintenance of the project :+1:

Regarding your comments: the @deprecate macro already exports the names for us. The tests that I've removed are not needed moving forward. They compare the result of the canonical method with the bang method, which is deprecated in future releases.