Merge pull request #40 from JuliaAI/dev

For a 0.4.2 release

Author: ablaom

.github/workflows/CI.yml

@@ -45,7 +45,11 @@ jobs:
             ${{ runner.os }}-
       - uses: julia-actions/julia-buildpkg@latest
       - uses: julia-actions/julia-runtest@latest
-      - uses: julia-actions/julia-uploadcodecov@v0.1
-        continue-on-error: true
+        env:
+          JULIA_NUM_THREADS: 2
+      - uses: julia-actions/julia-processcoverage@v1
+      - uses: codecov/codecov-action@v3
+        with:
+          file: lcov.info
 
 

Project.toml

@@ -1,7 +1,7 @@
 name = "MLJEnsembles"
 uuid = "50ed68f4-41fd-4504-931a-ed422449fee0"
 authors = ["Anthony D. Blaom <anthony.blaom@gmail.com>"]
-version = "0.4.1"
+version = "0.4.2"
 
 [deps]
 CategoricalArrays = "324d7699-5711-5eae-9e2f-1d82baa6b597"

README.md

@@ -4,11 +4,11 @@
 
 A package to create bagged homogeneous ensembles of
 machine learning models using the
-[MLJ](https://alan-turing-institute.github.io/MLJ.jl/dev/) machine
+[MLJ](https://JuliaAI.github.io/MLJ.jl/dev/) machine
 learning framework.
 
 For combining models in more general ways, see the [Composing
-Models](https://alan-turing-institute.github.io/MLJ.jl/dev/composing_models/#Composing-Models)
+Models](https://JuliaAI.github.io/MLJ.jl/dev/composing_models/#Composing-Models)
 section of the MLJ manual.
 
 
@@ -34,16 +34,16 @@ using MLJBase, MLJEnsembles
 In this case you will also need to load code defining an atomic model
 to ensemble. The easiest way to do this is run `Pkg.add("MLJModels");
 using MLJModels` and use the `@load` macro.  See the [Loading Model
-Code](https://alan-turing-institute.github.io/MLJ.jl/dev/loading_model_code/)
+Code](https://JuliaAI.github.io/MLJ.jl/dev/loading_model_code/)
 of the MLJ manual for this and other possibilities.
 
 
 ## Sample usage
 
-See [Data Science Tutorials](https://alan-turing-institute.github.io/DataScienceTutorials.jl/getting-started/ensembles/).
+See [Data Science Tutorials](https://JuliaAI.github.io/DataScienceTutorials.jl/getting-started/ensembles/).
 
 
 ## Documentation
 
-See the [MLJ manual](https://alan-turing-institute.github.io/MLJ.jl/dev/homogeneous_ensembles/#Homogeneous-Ensembles).
+See the [MLJ manual](https://JuliaAI.github.io/MLJ.jl/dev/homogeneous_ensembles/#Homogeneous-Ensembles).
 

src/ensembles.jl

@@ -412,7 +412,9 @@ end
 # Random._GLOBAL_RNG() and Random.default_rng() are threadsafe by default_rng
 # as they have thread local state from julia >=1.3<=1.6 and task local state Julia >=1.7   
 threadsafe_rng(rng::typeof(Random.default_rng())) = rng
-threadsafe_rng(rng::Random._GLOBAL_RNG) = rng
+if isdefined(Random, :_GLOBAL_RNG)
+    threadsafe_rng(rng::Random._GLOBAL_RNG) = rng
+end
 threadsafe_rng(rng) = deepcopy(rng)
 
 function _fit(res::CPUThreads, func, verbosity, stuff)

test/ensembles.jl

@@ -1,15 +1,3 @@
-module TestEnsembles
-
-using Test
-using Random
-using StableRNGs
-using MLJEnsembles
-using MLJBase
-using ..Models
-using CategoricalArrays
-import Distributions
-using StatisticalMeasures
-
 ## HELPER FUNCTIONS
 
 @test MLJEnsembles._reducer([1, 2], [3, ]) == [1, 2, 3]
@@ -187,10 +175,10 @@ predict(ensemble_model, fitresult, MLJEnsembles.selectrows(X, test))
 
 @testset "further test of sample weights" begin
     ## Note: This testset also indirectly tests for compatibility with the data-front end
-    # implemented by `KNNClassifier` as calls to `fit`/`predict` on an `Ensemble` model 
+    # implemented by `KNNClassifier` as calls to `fit`/`predict` on an `Ensemble` model
     # with `atom=KNNClassifier` would error if the ensemble implementation doesn't handle
     # data front-end conversions properly.
-    
+
     rng = StableRNG(123)
     N = 20
     X = (x = rand(rng, 3N), );
@@ -224,18 +212,18 @@ predict(ensemble_model, fitresult, MLJEnsembles.selectrows(X, test))
 end
 
 
-## MACHINE TEST 
+## MACHINE TEST
 ## (INCLUDES TEST OF UPDATE.
-## ALSO INCLUDES COMPATIBILITY TESTS FOR ENSEMBLES WITH ATOM MODELS HAVING A 
+## ALSO INCLUDES COMPATIBILITY TESTS FOR ENSEMBLES WITH ATOM MODELS HAVING A
 ## DIFFERENT DATA FRONT-END SEE #16)
 
-@testset "machine tests" begin
+@testset_accelerated "machine tests" acceleration begin
     N =100
     X = (x1=rand(N), x2=rand(N), x3=rand(N))
     y = 2X.x1  - X.x2 + 0.05*rand(N)
 
     atom = KNNRegressor(K=7)
-    ensemble_model = EnsembleModel(model=atom)
+    ensemble_model = EnsembleModel(; model=atom, acceleration)
     ensemble = machine(ensemble_model, X, y)
     train, test = partition(eachindex(y), 0.7)
     fit!(ensemble, rows=train, verbosity=0)
@@ -264,15 +252,13 @@ end
         atom;
         bagging_fraction=0.6,
         rng=123,
-        out_of_bag_measure = [log_loss, brier_score]
+        out_of_bag_measure = [log_loss, brier_score],
+        acceleration,
     )
     ensemble = machine(ensemble_model, X_, y_)
     fit!(ensemble)
     @test length(ensemble.fitresult.ensemble) == ensemble_model.n
 
 end
 
-
-end
-
 true

test/runtests.jl

@@ -1,4 +1,26 @@
-include("_models.jl")
+using Distributed
+# Thanks to https://stackoverflow.com/a/70895939/5056635 for the exeflags tip.
+addprocs(; exeflags="--project=$(Base.active_project())")
+
+@info "nprocs() = $(nprocs())"
+import .Threads
+@info "nthreads() = $(Threads.nthreads())"
+
+include("test_utilities.jl")
+include_everywhere("_models.jl")
+
+@everywhere begin
+    using Test
+    using Random
+    using StableRNGs
+    using MLJEnsembles
+    using MLJBase
+    using ..Models
+    using CategoricalArrays
+    import Distributions
+    using StatisticalMeasures
+    import Distributed
+end
+
 include("ensembles.jl")
 include("serialization.jl")
-

test/test_utilities.jl

@@ -0,0 +1,50 @@
+using Test
+
+using ComputationalResources
+
+macro testset_accelerated(name::String, var, ex)
+    testset_accelerated(name, var, ex)
+end
+macro testset_accelerated(name::String, var, opts::Expr, ex)
+    testset_accelerated(name, var, ex; eval(opts)...)
+end
+function testset_accelerated(name::String, var, ex; exclude=[])
+    final_ex = quote
+       local $var = CPU1()
+        @testset $name $ex
+    end
+
+    resources = AbstractResource[CPUProcesses(), CPUThreads()]
+
+    for res in resources
+        if any(x->typeof(res)<:x, exclude)
+            push!(final_ex.args, quote
+               local $var = $res
+               @testset $(name*" ($(typeof(res).name))") begin
+                   @test_broken false
+               end
+            end)
+        else
+            push!(final_ex.args, quote
+               local $var = $res
+               @testset $(name*" ($(typeof(res).name))") $ex
+            end)
+        end
+    end
+    # preserve outer location if possible
+    if ex isa Expr && ex.head === :block && !isempty(ex.args) &&
+        ex.args[1] isa LineNumberNode
+        final_ex = Expr(:block, ex.args[1], final_ex)
+    end
+    return esc(final_ex)
+end
+
+function include_everywhere(filepath)
+    include(filepath) # Load on Node 1 first, triggering any precompile
+    if nprocs() > 1
+        fullpath = joinpath(@__DIR__, filepath)
+        @sync for p in workers()
+            @async remotecall_wait(include, p, fullpath)
+        end
+    end
+end