Investigate loading time of LossFunctions.jl

[ablaom]

This used to be quite long, according to [this comment](See #77 (comment)). This seems surprising, given the modest deps:

LearnBase = "7f8f8fb0-2700-5f03-b4bd-41f8cfc144b6"
Markdown = "d6f4376e-aef5-505a-96c1-9c027394607a"
RecipesBase = "3cdcf5f2-1ef4-517c-9805-6587b60abb01"
StatsBase = "2913bbd2-ae8a-5f71-8c99-4fb6c76f3a91"

Worth reviewing.

[juliohm]

We dropped most deps in a recent refactoring of the code. LossFunctions.jl v0.9 only depends on CategoricalArrays.jl.

[ablaom]

Will check the timings again after resolving #896

[ablaom]

Based on PR at #898 :

julia> @time_imports using MLJBase
[ Info: Precompiling MLJBase [a7f614a8-145f-11e9-1d2a-a57a1082229d]
      0.2 ms  Reexport
      0.5 ms  ScientificTypesBase
      0.1 ms  DataValueInterfaces
      1.2 ms  DataAPI
      0.1 ms  IteratorInterfaceExtensions
      0.1 ms  TableTraits
     22.8 ms  Tables
      7.1 ms  Missings
      0.3 ms  Requires
    112.5 ms  CategoricalArrays 16.61% compilation time (6% recompilation)
     83.7 ms  FixedPointNumbers
     73.2 ms  ColorTypes 6.43% compilation time
      0.7 ms  Formatting
     75.3 ms  StringManipulation
    100.1 ms  Crayons
      0.9 ms  LaTeXStrings
    228.8 ms  PrettyTables
      0.8 ms  Compat
     75.7 ms  DataStructures
      0.4 ms  SortingAlgorithms
      2.5 ms  DocStringExtensions 67.38% compilation time
      6.7 ms  IrrationalConstants
     73.2 ms  ChainRulesCore
      0.7 ms  ChangesOfVariables
      1.1 ms  InverseFunctions
      0.7 ms  LogExpFunctions
      0.3 ms  StatsAPI
     23.1 ms  StatsBase
     23.0 ms  PDMats
      1.3 ms  OpenLibm_jll
     27.3 ms  Preferences
      0.3 ms  JLLWrappers
     10.4 ms  CompilerSupportLibraries_jll
     49.8 ms  OpenSpecFun_jll 98.31% compilation time (96% recompilation)
     16.4 ms  SpecialFunctions
      0.7 ms  Rmath_jll
     74.6 ms  Rmath 90.42% compilation time
      0.3 ms  NaNMath
      2.0 ms  Calculus
    107.8 ms  DualNumbers
      1.2 ms  HypergeometricFunctions
      7.1 ms  StatsFuns
      3.7 ms  QuadGK
    200.5 ms  FillArrays
      1.8 ms  DensityInterface
    254.7 ms  Distributions
    268.7 ms  ScientificTypes
      0.3 ms  StatisticalTraits
      4.2 ms  MLJModelInterface
      0.2 ms  UnPack
      0.5 ms  Parameters
      4.2 ms  InvertedIndices
      4.2 ms  ComputationalResources
      6.3 ms  ProgressMeter
     12.4 ms  LossFunctions   <-------------------------------------
      0.2 ms  SnoopPrecompile
      4.4 ms  StaticArraysCore
   1016.6 ms  StaticArrays
    325.8 ms  MarchingCubes
    908.5 ms  Colors
      0.3 ms  TensorCore
    299.7 ms  ColorVectorSpace 1.71% compilation time
     16.5 ms  ColorSchemes
      2.2 ms  Contour
   3430.1 ms  UnicodePlots 1.12% compilation time
     12.6 ms  CategoricalDistributions
    697.9 ms  MLJBase 16.62% compilation time

This looks reasonable.