Migration, standardizer and a few fixes

Project.toml

@@ -22,7 +22,7 @@ StatsBase = "2913bbd2-ae8a-5f71-8c99-4fb6c76f3a91"
 Tables = "bd369af6-aec1-5ad0-b16a-f7cc5008161c"
 
 [compat]
-MLJBase = "0.5"
+MLJBase = "0.6"
 julia = "1"
 
 [extras]

src/Clustering.jl

@@ -6,6 +6,7 @@
 module Clustering_
 
 import MLJBase
+import MLJBase: @mlj_model
 using ScientificTypes
 
 import ..Clustering # strange sytax for lazy-loading
@@ -15,8 +16,6 @@ using LinearAlgebra: norm
 
 const C = Clustering
 
-import ..@mlj_model
-
 const KMeansDescription =
     """
     K-Means algorithm: find K centroids corresponding to K clusters in the data.

src/MLJModels.jl

@@ -1,35 +1,25 @@
 module MLJModels
 
+using MLJBase, Tables, ScientificTypes
+using Requires, Pkg.TOML, OrderedCollections
+using StatsBase # countmap is required in metadata
+
 # for administrators to update Metadata.toml:
 export @update
 
-# from builtins/Transformers.jl:
-export StaticTransformer, FillImputer, FeatureSelector,
-    UnivariateStandardizer, Standardizer,
-    UnivariateBoxCoxTransformer,
-    OneHotEncoder
-
-# from builtins/Constant.jl:
-export ConstantRegressor, ConstantClassifier
-
-# from builtins/KNN.jl:
-export KNNRegressor
-
 # from loading.jl:
 export load, @load, info
 
 # from model_search:
 export models, localmodels
 
-using Requires
-using OrderedCollections
-using MLJBase
-using ScientificTypes
-using Tables
-using ColorTypes
-using StatsBase
+# from model/Constant
+export ConstantRegressor, ConstantClassifier,
+        DeterministicConstantRegressor, DeterministicConstantClassifier
 
-using Pkg.TOML
+# from model/Transformers
+export FeatureSelector, StaticTransformer, UnivariateStandardizer,
+        Standardizer, UnivariateBoxCoxTransformer, OneHotEncoder, FillImputer
 
 const srcdir = dirname(@__FILE__) # the directory containing this file
 
@@ -40,25 +30,32 @@ if VERSION < v"1.3"
 end
 nonmissing = nonmissingtype
 
-
-include("metadata_utils.jl")
-
 include("metadata.jl")
 include("model_search.jl")
 include("loading.jl")
 include("registry/src/Registry.jl")
 import .Registry.@update
 
 # load built-in models:
-include("builtins/Transformers.jl")
 include("builtins/Constant.jl")
 
-include("parameters_utils.jl")
-include("metadata_utils.jl")
+include("builtins/Transformers.jl")
+
+const INFO_GIVEN_HANDLE = Dict{Handle,Any}()
+const PKGS_GIVEN_NAME   = Dict{String,Vector{String}}()
+const AMBIGUOUS_NAMES   = String[]
+const NAMES             = String[]
 
+# lazily load in strap-on model interfaces for external packages:
 function __init__()
+    metadata_file = joinpath(srcdir, "registry", "Metadata.toml")
+
+    merge!(INFO_GIVEN_HANDLE, info_given_handle(metadata_file))
+    merge!(PKGS_GIVEN_NAME, pkgs_given_name(INFO_GIVEN_HANDLE))
+    append!(AMBIGUOUS_NAMES, ambiguous_names(INFO_GIVEN_HANDLE))
+    append!(NAMES, model_names(INFO_GIVEN_HANDLE))
+    @info "Model metadata loaded from registry. "
 
-    # lazily load in strap-on model interfaces for external packages:
     @require MultivariateStats="6f286f6a-111f-5878-ab1e-185364afe411" include("MultivariateStats.jl")
     @require DecisionTree="7806a523-6efd-50cb-b5f6-3fa6f1930dbb" include("DecisionTree.jl")
     @require GaussianProcesses="891a1506-143c-57d2-908e-e1f8e92e6de9" include("GaussianProcesses.jl")
@@ -69,7 +66,6 @@ function __init__()
     @require XGBoost = "009559a3-9522-5dbb-924b-0b6ed2b22bb9" include("XGBoost.jl")
     @require LIBSVM="b1bec4e5-fd48-53fe-b0cb-9723c09d164b" include("LIBSVM.jl")
     @require NearestNeighbors="b8a86587-4115-5ab1-83bc-aa920d37bbce" include("NearestNeighbors.jl")
-
 end
 
 end # module

src/MultivariateStats.jl

@@ -3,6 +3,7 @@ module MultivariateStats_
 export RidgeRegressor, PCA, KernelPCA, ICA
 
 import MLJBase
+import MLJBase: @mlj_model
 using ScientificTypes
 using Tables
 
@@ -15,8 +16,6 @@ struct LinearFitresult{F} <: MLJBase.MLJType
     bias::F
 end
 
-import ..@mlj_model
-
 ####
 #### RIDGE
 ####

src/NearestNeighbors.jl

@@ -1,10 +1,10 @@
 module NearestNeighbors_
 
 import MLJBase
+import MLJBase: @mlj_model, metadata_model, metadata_pkg
 using Distances
 
 import ..NearestNeighbors
-import ..@mlj_model, ..metadata_pkg, ..metadata_model
 
 const NN = NearestNeighbors
 
@@ -70,32 +70,6 @@ end
 
 const KNN = Union{KNNRegressor, KNNClassifier}
 
-function MLJBase.clean!(m::KNN)
-    warning = ""
-    if m.K < 1
-        warning *= "Number of neighbors 'K' needs to be larger than 0. Setting to 1.\n"
-        m.K = 1
-    end
-    if m.leafsize < 0
-        warning *= "Leaf size should be ≥ 0. Setting to 10.\n"
-        m.leafsize = 10
-    end
-    if m.algorithm ∉ (:kdtree, :brutetree, :balltree)
-        warning *= "The tree algorithm should be ':kdtree', ':brutetree' or ':balltree'." *
-                   "Setting to ':kdtree'.\n"
-        m.algorithm = :kdtree
-    end
-    if m.algorithm == :kdtree && !isa(m.metric ∉ (Euclidean, Chebyshev, Minkowski, Citiblock))
-        warning *= "KDTree only supports axis-aligned metrics. Setting to 'Euclidean'.\n"
-        m.metric = Euclidean()
-    end
-    if m.weights ∉ (:uniform, :distance)
-        warning *= "Weighing should be ':uniform' or ':distance'. Setting to ':uniform'.\n"
-        m.weights = :distance
-    end
-    return warning
-end
-
 function MLJBase.fit(m::KNN, verbosity::Int, X, y)
     Xmatrix = MLJBase.matrix(X, transpose=true) # NOTE: copies the data
     if m.algorithm == :kdtree

src/ScikitLearn/ScikitLearn.jl

@@ -2,6 +2,8 @@ module ScikitLearn_
 
 #> for all Supervised models:
 import MLJBase
+import MLJBase: @mlj_model, metadata_model,
+				_process_model_def, _model_constructor, _model_cleaner
 using ScientificTypes
 using Tables
 
@@ -23,9 +25,6 @@ import ..ScikitLearn
 
 include("svm.jl")
 
-import .._process_model_def, .._model_constructor, .._model_cleaner
-import  ..metadata_model # metadata_pkg is handled by @sk_model
-
 const Option{T} = Union{Nothing, T}
 
 const SKLM = ((ScikitLearn.Skcore).pyimport("sklearn.linear_model"))

src/builtins/Constant.jl

@@ -1,17 +1,10 @@
-# this file defines *and* loads one module
-
 module Constant
 
-export ConstantRegressor, ConstantClassifier
-export DeterministicConstantRegressor, DeterministicConstantClassifier
+using ..MLJBase, ..Tables
+using Distributions
 
-import MLJBase
-import MLJBase.nrows
-import Distributions
-using StatsBase
-using Statistics
-using CategoricalArrays
-using ScientificTypes
+export ConstantRegressor, ConstantClassifier,
+        DeterministicConstantRegressor, DeterministicConstantClassifier
 
 ## THE CONSTANT REGRESSOR
 
@@ -23,7 +16,7 @@ probability distribution best fitting the training target data. Use
 `predict_mean` to predict the mean value instead.
 
 """
-struct ConstantRegressor{D} <: MLJBase.Probabilistic 
+struct ConstantRegressor{D} <: MLJBase.Probabilistic
     distribution_type::Type{D}
 end
 function ConstantRegressor(; distribution_type=Distributions.Normal)
@@ -62,8 +55,8 @@ MLJBase.package_uuid(::Type{<:ConstantRegressor}) =
     "d491faf4-2d78-11e9-2867-c94bc002c0b7"
 MLJBase.package_url(::Type{<:ConstantRegressor}) = "https://github.com/alan-turing-institute/MLJModels.jl"
 MLJBase.is_pure_julia(::Type{<:ConstantRegressor}) = true
-MLJBase.input_scitype(::Type{<:ConstantRegressor}) = Table(Scientific) # anything goes
-MLJBase.target_scitype(::Type{<:ConstantRegressor}) = AbstractVector{Continuous}
+MLJBase.input_scitype(::Type{<:ConstantRegressor}) = MLJBase.Table(MLJBase.Scientific) # anything goes
+MLJBase.target_scitype(::Type{<:ConstantRegressor}) = AbstractVector{MLJBase.Continuous}
 
 
 ## THE CONSTANT DETERMINISTIC REGRESSOR (FOR TESTING)
@@ -85,8 +78,8 @@ MLJBase.package_name(::Type{<:DeterministicConstantRegressor}) = MLJBase.package
 MLJBase.package_uuid(::Type{<:DeterministicConstantRegressor}) = MLJBase.package_url(ConstantRegressor)
 MLJBase.package_url(::Type{<:DeterministicConstantRegressor}) = MLJBase.package_url(ConstantRegressor)
 MLJBase.is_pure_julia(::Type{<:DeterministicConstantRegressor}) = true
-MLJBase.input_scitype(::Type{<:DeterministicConstantRegressor}) = Table(Scientific) # anything goes
-MLJBase.target_scitype(::Type{<:DeterministicConstantRegressor}) = AbstractVector{Continuous}
+MLJBase.input_scitype(::Type{<:DeterministicConstantRegressor}) = MLJBase.Table(MLJBase.Scientific) # anything goes
+MLJBase.target_scitype(::Type{<:DeterministicConstantRegressor}) = AbstractVector{MLJBase.Continuous}
 
 
 ## THE CONSTANT CLASSIFIER
@@ -104,7 +97,7 @@ obtain the training target mode instead.
 struct ConstantClassifier <: MLJBase.Probabilistic end
 
 function MLJBase.fit(model::ConstantClassifier,
-                 verbosity::Int, X, y) 
+                 verbosity::Int, X, y)
 
     fitresult = Distributions.fit(MLJBase.UnivariateFinite, y)
 
@@ -127,8 +120,8 @@ MLJBase.package_name(::Type{<:ConstantClassifier}) = MLJBase.package_name(Consta
 MLJBase.package_uuid(::Type{<:ConstantClassifier}) = MLJBase.package_uuid(ConstantRegressor)
 MLJBase.package_url(::Type{<:ConstantClassifier}) = MLJBase.package_url(ConstantRegressor)
 MLJBase.is_pure_julia(::Type{<:ConstantClassifier}) = true
-MLJBase.input_scitype(::Type{<:ConstantClassifier}) = Table(Scientific) # anything goes
-MLJBase.target_scitype(::Type{<:ConstantClassifier}) = AbstractVector{<:Finite}
+MLJBase.input_scitype(::Type{<:ConstantClassifier}) = MLJBase.Table(MLJBase.Scientific) # anything goes
+MLJBase.target_scitype(::Type{<:ConstantClassifier}) = AbstractVector{<:MLJBase.Finite}
 
 
 ## DETERMINISTIC CONSTANT CLASSIFIER (FOR TESTING)
@@ -161,13 +154,9 @@ MLJBase.package_name(::Type{<:DeterministicConstantClassifier}) = MLJBase.packag
 MLJBase.package_uuid(::Type{<:DeterministicConstantClassifier}) = MLJBase.package_uuid(ConstantRegressor)
 MLJBase.package_url(::Type{<:DeterministicConstantClassifier}) = MLJBase.package_url(ConstantRegressor)
 MLJBase.is_pure_julia(::Type{<:DeterministicConstantClassifier}) = true
-MLJBase.input_scitype(::Type{<:DeterministicConstantClassifier}) = Table(Scientific) # anything goes
-MLJBase.target_scitype(::Type{<:DeterministicConstantClassifier}) = AbstractVector{<:Finite}
-
+MLJBase.input_scitype(::Type{<:DeterministicConstantClassifier}) = MLJBase.Table(MLJBase.Scientific) # anything goes
+MLJBase.target_scitype(::Type{<:DeterministicConstantClassifier}) = AbstractVector{<:MLJBase.Finite}
 
-end # module
-
-
-## EXPOSE THE INTERFACE
+end
 
-using .Constant
+using MLJModels.Constant