add final trait tests

Author: ablaom

Project.toml

@@ -12,6 +12,8 @@ LinearAlgebra = "37e2e46d-f89d-539d-b4ee-838fcccc9c8e"
 MLUtils = "f1d291b0-491e-4a28-83b9-f70985020b54"
 MacroTools = "1914dd2f-81c6-5fcd-8719-6d5c9610ff09"
 Random = "9a3f8284-a2c9-5f02-9a11-845980a1fd5c"
+ScientificTypes = "321657f4-b219-11e9-178b-2701a2544e81"
+ScientificTypesBase = "30f210dd-8aff-4c5f-94ba-8e64358c1161"
 Serialization = "9e88b42a-f829-5b0c-bbe9-9e923198166b"
 StableRNGs = "860ef19b-820b-49d6-a774-d7a799459cd3"
 Statistics = "10745b16-79ce-11e8-11f9-7d13ad32a3b2"
@@ -27,6 +29,8 @@ LearnAPI = "0.1.0"
 LinearAlgebra = "<0.0.1, 1"
 MLUtils = "0.4"
 Random = "<0.0.1, 1"
+ScientificTypes = "3.0.2"
+ScientificTypesBase = "3.0.0"
 Serialization = "<0.0.1, 1"
 StableRNGs = "1"
 Statistics = "<0.0.1, 1"

src/LearnTestAPI.jl

@@ -34,6 +34,17 @@ import StableRNGs
 import InteractiveUtils
 import MacroTools
 import IsURL
+import ScientificTypes
+
+# duplication of import/using statements in files at src/learners not appearing above, for
+# test learners:
+import Distributions
+using ScientificTypesBase
+using Tables
+using LinearAlgebra
+using Random
+using Statistics
+using UnPack
 
 include("tools.jl")
 include("logging.jl")

src/learners/ensembling.jl

@@ -8,7 +8,7 @@ using LinearAlgebra
 using Random
 using Statistics
 using UnPack
-
+using ScientificTypesBase
 
 # # ENSEMBLE OF REGRESSORS (A MODEL WRAPPER)
 
@@ -542,6 +542,8 @@ end
     iteration_parameter = :ntrees,
     kinds_of_proxy = (Point(),),
     tags = ("regression", "ensembling", "iterative algorithms"),
+    fit_scitype = Tuple{AbstractVector{Continuous},AbstractVector{Continuous}},
+    target_observation_scitype = Continuous,
     functions = (
         :(LearnAPI.fit),
         :(LearnAPI.learner),

src/logging.jl

@@ -402,12 +402,23 @@ const HUMAN_NAME = """
     Checking that `LearnAPI.human_name(learner)` is a string.
 
   """
+const FIT_SCITYPE = """
 
-function MISSING_TRAITS(missing_traits)
+    Checking that `data` supplied for testing satisfies the constraints articulated by
+    `LearnAPI.fit_scitype(learner)`.
+
+  """
+const TARGET_OBSERVATION_SCITYPE = """
+
+    Checking that `LearnAPI.target(learner, observatoins)` satisifies the constraints
+    articulated by `LearnAPI.target_observation_scitype(learner)`.
+
+"""
+function MISSING_TRAITS(missing_traits, name)
     list = join( map(ex->"`$ex`",collect(missing_traits)), ", ", " and ")
     return """
 
-    The following optional traits have not been overloaded:
+    The following traits, deemed optional for $name, have not been overloaded:
 
     $list.
 

src/testapi.jl

@@ -1,5 +1,5 @@
 """
-    @testapi learner dataset1, dataset2 ... verbosity=1
+    @testapi learner dataset1 dataset2 ... verbosity=1
 
 Test that `learner` correctly implements the LearnAPI.jl interface, by checking
 contracts against one or more data sets.
@@ -37,15 +37,20 @@ The following are *not* tested:
 
 - That the output of `LearnAPI.target(learner, data)` is indeed a target, in the sense
   that it can be paired, in some way, with the output of `predict`. Such a test would be
-  to suitably pair the output with a predicted proxy for the target, using, say, a proper
-  scoring rule, in the case of probabilistic predictions.
+  to suitably pair the output with a predicted proxy for the target, using, for example, a
+  proper scoring rule, in the case of probabilistic predictions.
 
 - That `inverse_transform` is an approximate left or right inverse to `transform`
 
 - That the one-line convenience methods, `transform(learner, ...)` or `predict(learner,
   ...)`, where implemented, have the same effect as the two-line calls they combine.
 
-- The veracity of `is_pure_julia(learner)`.
+- The veracity of `LearnAPI.is_pure_julia(learner)`.
+
+- The second of the two contracts appearing in the
+  [`LearnAPI.target_observation_scitype`](@extref) docstring. The first contract is only
+  tested if `LearnAPI.data_interface(learner)` is `LearnAPI.RandomAccess()` or
+  `LearnAPI.FiniteIterable()`.
 
 Whenever the internal `learner` algorithm involves case distinctions around data or
 hyperparameters, it is recommended that multiple datasets, and learners with a variety of
@@ -637,12 +642,34 @@ macro testapi(learner, data...)
             end
 
             @logged_testset $HUMAN_NAME verbosity begin
-                human_name = LearnAPI.human_name(learner)
-                Test.@test human_name isa String
+                Test.@test _human_name isa String
+            end
+
+            @logged_testset $FIT_SCITYPE verbosity begin
+                S = LearnAPI.fit_scitype(learner)
+                if S == Union{}
+                    push!(missing_traits, :(LearnAPI.fit_scitype))
+                else
+                    Test.@test ScientificTypes.scitype(data) <: S
+                end
+            end
+
+            S = LearnAPI.target_observation_scitype(learner)
+            testable = :(LearnAPI.target) in _functions &&
+               _data_interface in (LearnAPI.RandomAccess(), LearnAPI.FiniteIterable())
+            if S == Any
+                push!(missing_traits, :(LearnAPI.target_observation_scitype))
+            elseif testable
+                @logged_testset $TARGET_OBSERVATION_SCITYPE verbosity begin
+                    Test.@test all([o for o in _y]) do o
+                        ScientificTypes.scitype(o) <: S
+                    end
+                end
             end
 
         end # for loop over datasets
-        verbosity > 0 && !isempty(missing_traits) && @info $MISSING_TRAITS(missing_traits)
+        verbosity > 0 && !isempty(missing_traits) &&
+            @info $MISSING_TRAITS(missing_traits, _human_name)
         verbosity > 0 && @info "------ @testapi - $_human_name - tests complete ------"
         nothing
     end # quote