Merge pull request #163 from JuliaAI/dev

For a 1.7.0 release

.github/workflows/ci.yml

@@ -17,7 +17,6 @@ jobs:
       fail-fast: false
       matrix:
         version:
-          - '1.0'
           - '1.6'
           - '1'
         os:

Project.toml

@@ -1,7 +1,7 @@
 name = "MLJModelInterface"
 uuid = "e80e1ace-859a-464e-9ed9-23947d8ae3ea"
 authors = ["Thibaut Lienart and Anthony Blaom"]
-version = "1.6.0"
+version = "1.7.0"
 
 [deps]
 Random = "9a3f8284-a2c9-5f02-9a11-845980a1fd5c"
@@ -11,17 +11,18 @@ StatisticalTraits = "64bff920-2084-43da-a3e6-9bb72801c0c9"
 [compat]
 ScientificTypesBase = "3.0"
 StatisticalTraits = "3.2"
-julia = "1"
+julia = "1.6"
 
 [extras]
 CategoricalArrays = "324d7699-5711-5eae-9e2f-1d82baa6b597"
 DataFrames = "a93c6f00-e57d-5684-b7b6-d8193f3e46c0"
 Distances = "b4f34e82-e78d-54a5-968a-f98e89d6e8f7"
 InteractiveUtils = "b77e0a4c-d291-57a0-90e8-8db25a27a240"
 Markdown = "d6f4376e-aef5-505a-96c1-9c027394607a"
+OrderedCollections = "bac558e1-5e72-5ebc-8fee-abe8a469f55d"
 ScientificTypes = "321657f4-b219-11e9-178b-2701a2544e81"
 Tables = "bd369af6-aec1-5ad0-b16a-f7cc5008161c"
 Test = "8dfed614-e22c-5e08-85e1-65c5234f0b40"
 
 [targets]
-test = ["CategoricalArrays", "DataFrames", "Distances", "InteractiveUtils", "Markdown", "ScientificTypes", "Tables", "Test"]
+test = ["CategoricalArrays", "DataFrames", "Distances", "InteractiveUtils", "Markdown", "OrderedCollections", "ScientificTypes", "Tables", "Test"]

src/metadata_utils.jl

@@ -65,11 +65,16 @@ function _extend!(program::Expr, trait::Symbol, value, T)
     end
 end
 
-const DEPWARN_DOCSTRING =
-    "`metadata_model` should not be called with the keyword argument "*
-    "`descr` or `docstring`. Implementers of the MLJ model interface "*
-    "should instead create an MLJ-compliant docstring in the usual way. "*
-    "See https://alan-turing-institute.github.io/MLJ.jl/dev/adding_models_for_general_use/#Document-strings for details. "
+depwarn_docstring(T) =
+    """
+
+    Regarding $T: `metadata_model` should not be called with the keyword argument `descr`
+    or `docstring`. Implementers of the MLJ model interface should instead create an
+    MLJ-compliant docstring in the usual way.  See
+    https://alan-turing-institute.github.io/MLJ.jl/dev/adding_models_for_general_use/#Document-strings
+    for details.
+
+    """
 
 """
     metadata_model(T; args...)
@@ -122,7 +127,7 @@ function metadata_model(
     supports_training_losses::Union{Nothing,Bool}=nothing,
     reports_feature_importances::Union{Nothing,Bool}=nothing,
 )
-    docstring === nothing || Base.depwarn(DEPWARN_DOCSTRING, :metadata_model)
+    docstring === nothing || Base.depwarn(depwarn_docstring(T), :metadata_model)
 
     program = quote end
 

src/model_api.jl

@@ -100,7 +100,7 @@ part of the tuple returned by `fit`.
 
 """
 fitted_params(::Model, fitresult) = (fitresult=fitresult,)
-
+fitted_params(::Static, ::Nothing) = nothing
 """
 
     predict(model, fitresult, new_data...)
@@ -173,6 +173,8 @@ the feature importances from the model's `fitresult` and `report` as an
 abstract vector of `feature::Symbol => importance::Real` pairs
 (e.g `[:gender =>0.23, :height =>0.7, :weight => 0.1]`).
 
+# New model implementations
+
 The following trait overload is also required:
 `MLJModelInterface.reports_feature_importances(::Type{<:M}) = true`
 
@@ -182,3 +184,56 @@ If for some reason a model is sometimes unable to report feature importances the
 
 """
 function feature_importances end
+
+_named_tuple(named_tuple::NamedTuple) = named_tuple
+_named_tuple(::Nothing) = NamedTuple()
+_named_tuple(something_else) = (report=something_else,)
+_scrub(x) = x
+_scrub(x::NamedTuple) = isempty(x) ? nothing : x
+_keys(named_tuple) = keys(named_tuple)
+_keys(::Nothing) = ()
+
+"""
+    MLJModelInterface.report(model, report_given_method)
+
+Merge the reports in the dictionary `report_given_method` into a single
+property-accessible object. It is supposed that each key of the dictionary is either
+`:fit` or the name of an operation, such as `:predict` or `:transform`. Each value will be
+the `report` component returned by a training method (`fit` or `update`) dispatched on the
+`model` type, in the case of `:fit`, or the report component returned by an operation that
+supports reporting.
+
+# New model implementations
+
+Overloading this method is optional, unless the model generates reports that are neither
+named tuples nor `nothing`.
+
+Assuming each value in the `report_given_method` dictionary is either a named tuple 
+or `nothing`, and there are no conflicts between the keys of the dictionary values 
+(the individual reports), the fallback returns the usual named tuple merge of the 
+dictionary values, ignoring any `nothing` value. If there is a key conflict, all operation 
+reports are first wrapped in a named
+tuple of length one, as in `(predict=predict_report,)`. A `:fit` report is never wrapped.
+
+If any dictionary `value` is neither a named tuple nor `nothing`, it is first wrapped as
+`(report=value, )` before merging.
+
+"""
+function report(model, report_given_method)
+
+    return_keys = vcat(collect.(_keys.(values(report_given_method)))...)
+
+    # Note that we want to avoid copying values in each individual report named tuple, and
+    # merge the reports in a reproducible order.
+
+    methods = collect(keys(report_given_method)) |> sort!
+    length(methods) == 1 && return _scrub(report_given_method[only(methods)])
+    need_to_wrap = return_keys != unique(return_keys)
+    reports = map(methods) do method
+        tup = _named_tuple(report_given_method[method])
+        isempty(tup) ? NamedTuple() :
+            (need_to_wrap && method !== :fit) ? NamedTuple{(method,)}((tup,)) :
+                tup
+    end
+    return _scrub(merge(reports...))
+end

test/data_utils.jl

@@ -323,9 +323,9 @@ end
     eval(:(module UserSide
             import MLJModelInterface: metadata_model, metadata_pkg
             struct A end
-            descr = "something"
+            human_name = "Big Foot"
             # Smoke tests.
-            metadata_model(A; descr=descr)
+            metadata_model(A; human_name)
             metadata_pkg(A)
     end))
 end

test/model_api.jl

@@ -5,7 +5,6 @@ end
     f0::Int
 end
 
-
 mutable struct APIx1 <: Static end
 
 @testset "selectrows(model, data...)" begin
@@ -95,3 +94,47 @@ mutable struct UnivariateFiniteFitter <: Probabilistic end
     @test yhat == fill(DummyUnivariateFinite(), 3)
 
 end
+
+@testset "fallback for `report()` method" begin
+    report_given_method =
+        OrderedCollections.OrderedDict(
+            :predict=>(y=7,),
+            :fit=>(x=1, z=3),
+            :transform=>nothing,
+        )
+    @test MLJModelInterface.report(APIx0(f0=1), report_given_method) ==
+        (x=1, z=3, y=7)
+
+    report_given_method =
+        OrderedCollections.OrderedDict(
+            :predict=>(y=7,),
+            :fit=>(y=1, z=3),
+            :transform=>nothing,
+        )
+    @test MLJModelInterface.report(APIx0(f0=1), report_given_method) ==
+        (y=1, z=3, predict=(y=7,))
+
+    @test MLJModelInterface.report(
+        APIx0(f0=1),
+        OrderedCollections.OrderedDict(:fit => nothing, :transform => NamedTuple()),
+    ) |> isnothing
+
+    @test MLJModelInterface.report(
+        APIx0(f0=1),
+        OrderedCollections.OrderedDict(:fit => 42),
+    ) == 42
+
+    @test MLJModelInterface.report(
+        APIx0(f0=1),
+        OrderedCollections.OrderedDict(:fit => nothing),
+    ) |> isnothing
+
+    @test MLJModelInterface.report(
+        APIx0(f0=1),
+        OrderedCollections.OrderedDict(:fit => NamedTuple()),
+    ) |> isnothing
+
+
+end
+
+

test/runtests.jl

@@ -3,6 +3,7 @@ using ScientificTypesBase, ScientificTypes
 using Tables, Distances, CategoricalArrays, InteractiveUtils
 import DataFrames: DataFrame
 import Markdown
+import OrderedCollections
 
 const M  = MLJModelInterface
 const FI = M.FullInterface