Merge pull request #118 from tylerjthomas9/master

Libstdcxx patch, remove load_boston examples

.github/workflows/ci.yml

@@ -29,24 +29,12 @@ jobs:
         julia-arch:
           - x64
     steps:
-      - uses: actions/checkout@v2
+      - uses: actions/checkout@v3
       - uses: julia-actions/setup-julia@v1
         with:
           version: ${{ matrix.julia-version }}
           arch: ${{ matrix.julia-arch }}
-      - name: Cache artifacts
-        uses: actions/cache@v2
-        env:
-          cache-name: cache-artifacts
-        with:
-          path: |
-            ~/.julia/artifacts
-            ~/.julia/registries
-          key: ${{ runner.os }}-test-${{ env.cache-name }}-${{ hashFiles('**/Project.toml') }}
-          restore-keys: |
-            ${{ runner.os }}-test-${{ env.cache-name }}-
-            ${{ runner.os }}-test-
-            ${{ runner.os }}-
+      - uses: julia-actions/cache@v1 # https://github.com/julia-actions/cache
       - uses: julia-actions/julia-buildpkg@latest
       - uses: julia-actions/julia-runtest@latest
       - uses: julia-actions/[email protected]
@@ -57,23 +45,13 @@ jobs:
   Documentation:
     runs-on: ubuntu-latest
     steps:
-      - uses: actions/checkout@v2
-      - uses: julia-actions/setup-julia@latest
-        with:
-          version: 1
-      - name: Cache artifacts
-        uses: actions/cache@v2
-        env:
-          cache-name: cache-artifacts
+      - uses: actions/checkout@v3
+      - uses: julia-actions/setup-julia@v1
         with:
-          path: |
-            ~/.julia/artifacts
-            ~/.julia/registries
-          key: ${{ runner.os }}-test-${{ env.cache-name }}-${{ hashFiles('**/docs/Project.toml') }}
-          restore-keys: |
-            ${{ runner.os }}-test-${{ env.cache-name }}-
-            ${{ runner.os }}-test-
-            ${{ runner.os }}-
+          version: 1.6
+          arch: x64
+      - uses: julia-actions/cache@v1 # https://github.com/julia-actions/cache
+      - uses: julia-actions/julia-buildpkg@latest
       - uses: julia-actions/julia-docdeploy@releases/v1
         env:
           DOCUMENTER_KEY: ${{ secrets.DOCUMENTER_KEY }}

Project.toml

@@ -1,6 +1,6 @@
 name = "ScikitLearn"
 uuid = "3646fa90-6ef7-5e7e-9f22-8aca16db6324"
-version = "0.6.5"
+version = "0.6.6"
 
 [deps]
 Compat = "34da2185-b29b-5c13-b0c7-acf172513d20"
@@ -21,12 +21,12 @@ VersionParsing = "81def892-9a0e-5fdd-b105-ffc91e053289"
 
 [compat]
 Compat = "2.2, 3, 4"
-Conda = "1.5.2"
+Conda = "1.5.2, 1.6, 1.7"
 DataFrames = "0.20, 0.21, 0.22, 1"
 IterTools = "1.2, 1.3"
 MacroTools = "0.5"
 Parameters = "0.12"
-PyCall = "1.92"
+PyCall = "1.92, 1.93, 1.94"
 ScikitLearnBase = "0.5"
 StatsBase = "0.33"
 VersionParsing = "1.2"

docs/Project.toml

@@ -13,12 +13,12 @@ Random = "9a3f8284-a2c9-5f02-9a11-845980a1fd5c"
 Statistics = "10745b16-79ce-11e8-11f9-7d13ad32a3b2"
 
 [compat]
-DataFrames = "0.21"
+DataFrames = "1"
 Documenter = "0.24"
 GaussianMixtures = "0.3"
 GaussianProcesses = "0.12"
 JLD = "0.9, 0.10"
-PyCall = "1.91"
+PyCall = "1.92"
 PyCallJLD = "0.2"
-PyPlot = "2.9"
-RDatasets = "0.6"
+PyPlot = ">=2.9"
+RDatasets = ">=0.6"

docs/make.jl

@@ -1,19 +1,11 @@
 using Conda
-if Base.VERSION <= v"1.6.2" 
-	# GLIBCXX_3.4.26 
-	Conda.add("libstdcxx-ng>=3.4,<9.2", channel="conda-forge")
-else 
-	# GLIBCXX_3.4.29 
-	# checked up to v1.8.0 
-	Conda.add("libstdcxx-ng>=3.4,<11.4", channel="conda-forge")
-end 
 
 using Pkg
 using Documenter
 using ScikitLearn
+ScikitLearn.Skcore.import_sklearn() #preinstalls scikit-learn
 using PyPlot #pre-installs matplotlib
 import PyCall
-PyCall.pyimport_conda("sklearn", "scikit-learn") #preinstalls scikit-learn
 
 
 pages = [

examples/Cross_Validated_Predictions.ipynb

@@ -23,16 +23,26 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 1,
+   "execution_count": null,
    "metadata": {
     "collapsed": false
    },
-   "outputs": [],
+   "outputs": [
+    {
+     "ename": "",
+     "evalue": "",
+     "output_type": "error",
+     "traceback": [
+      "\u001b[1;31mFailed to start the Kernel. \n",
+      "\u001b[1;31mView Jupyter <a href='command:jupyter.viewOutput'>log</a> for further details."
+     ]
+    }
+   ],
    "source": [
     "using ScikitLearn\n",
     "using ScikitLearn.CrossValidation: cross_val_predict\n",
     "\n",
-    "@sk_import datasets: load_boston\n",
+    "@sk_import datasets: fetch_california_housing\n",
     "@sk_import linear_model: LinearRegression\n",
     "using PyPlot"
    ]
@@ -57,12 +67,12 @@
    ],
    "source": [
     "lr = LinearRegression()\n",
-    "boston = load_boston()\n",
-    "y = boston[\"target\"]\n",
+    "data = fetch_california_housing()\n",
+    "y = data[\"target\"]\n",
     "\n",
     "# cross_val_predict returns an array of the same size as `y` where each entry\n",
     "# is a prediction obtained by cross validated:\n",
-    "predicted = cross_val_predict(lr, boston[\"data\"], y, cv=10)\n",
+    "predicted = cross_val_predict(lr, data[\"data\"], y, cv=10)\n",
     "\n",
     "scatter(y, predicted)\n",
     "plot([minimum(y), maximum(y)], [minimum(y), maximum(y)], \"k--\", lw=4)\n",
@@ -73,9 +83,9 @@
  ],
  "metadata": {
   "kernelspec": {
-   "display_name": "Julia 0.4.5",
+   "display_name": "Julia 1.7.3",
    "language": "julia",
-   "name": "julia-0.4"
+   "name": "julia-1.7"
   },
   "language_info": {
    "file_extension": ".jl",

src/Skcore.jl

@@ -115,56 +115,103 @@ symbols_in(e::Expr) = union(symbols_in(e.head), map(symbols_in, e.args)...)
 symbols_in(e::Symbol) = Set([e])
 symbols_in(::Any) = Set()
 
+# https://github.com/cjdoris/CondaPkg.jl/blob/main/src/resolve.jl#L77
+const _compatible_libstdcxx_ng_versions = [
+    (v"3.4.31", ">=3.4,<=13.1"),
+    (v"3.4.30", ">=3.4,<13.0"),
+    (v"3.4.29", ">=3.4,<12.0"),
+    (v"3.4.28", ">=3.4,<11.0"),
+    (v"3.4.27", ">=3.4,<9.3"),
+    (v"3.4.26", ">=3.4,<9.2"),
+    (v"3.4.25", ">=3.4,<9.0"),
+    (v"3.4.24", ">=3.4,<8.0"),
+    (v"3.4.23", ">=3.4,<7.2"),
+    (v"3.4.22", ">=3.4,<7.0"),
+    (v"3.4.21", ">=3.4,<6.0"),
+    (v"3.4.20", ">=3.4,<5.0"),
+    (v"3.4.19", ">=3.4,<4.9"),
+]
+
+"""
+    _compatible_libstdcxx_ng_version()
+
+Version of libstdcxx-ng compatible with the libstdc++ loaded into Julia.
+Specifying the package "libstdcxx-ng" with version "<=julia" will replace the version with
+this one. This should be used by anything which embeds Python into the Julia process - for
+instance it is used by PythonCall.
+"""
+function _compatible_libstdcxx_ng_version()
+    if !Sys.islinux()
+        return
+    end
+    # bound = get(ENV, "JULIA_CONDAPKG_LIBSTDCXX_VERSION_BOUND", "")
+    # if bound != ""
+    #     return bound
+    # end
+    loaded_libstdcxx_version = Base.BinaryPlatforms.detect_libstdcxx_version()
+    if loaded_libstdcxx_version === nothing
+        return 
+    end
+    for (version, bound) in _compatible_libstdcxx_ng_versions
+        if loaded_libstdcxx_version ≥ version
+            return bound
+        end
+    end
+    return "" # no bound for unknown libstcxx versions
+end
 
 mkl_checked= false #neccessary for hack
+libstdcxx_solved = false # skip libstdcxx install if it was already done
 function import_sklearn()
     global mkl_checked
+    global libstdcxx_solved
 
     @static if Sys.isapple()
       mod = try
-     #
-               if PyCall.conda && !mkl_checked
-                 try
-                   # check for existence of mkl-service. 
-                   # Numpy, sklearn, etc. requires either `mkl` or `no-mkl` service to run
-                   # By default Conda comes with mkl
-                   # For this package to run on MacOS the `no-mkl` versions of Numpy, sklearn is needed   
-                   pyimport("mkl")
+            if PyCall.conda && !mkl_checked
+                try
+                    # check for existence of mkl-service. 
+                    # Numpy, sklearn, etc. requires either `mkl` or `no-mkl` service to run
+                    # By default Conda comes with mkl
+                    # For this package to run on MacOS the `no-mkl` versions of Numpy, sklearn is needed   
+                    pyimport("mkl")
 
-                   #following Code runs only if mkl-service exists otherwise jumps to catch branch
-                   @info "Installing non-mkl versions of sci-kit learn via Conda"
-                   #use non-mkl versions of python packages when ENV["PYTHON"]="Conda" or "" is used
-                   #when a different non-conda local python is used everthing works fine
-                   Conda.add("nomkl")
-                   Conda.rm("mkl")#This also removes mkl-service
-                   #force reinstall of scikit-learn replacing any previous mkl version
-                   Conda.add("scikit-learn")
-                   Conda.add("openblas")
-                   Conda.add("llvm-openmp", channel = "conda-forge")
-                   mkl_checked = true
+                    #following Code runs only if mkl-service exists otherwise jumps to catch branch
+                    @info "Installing non-mkl versions of sci-kit learn via Conda"
+                    #use non-mkl versions of python packages when ENV["PYTHON"]="Conda" or "" is used
+                    #when a different non-conda local python is used everthing works fine
+                    Conda.add("nomkl")
+                    Conda.rm("mkl")#This also removes mkl-service
+                    #force reinstall of scikit-learn replacing any previous mkl version
+                    Conda.add("scikit-learn")
+                    Conda.add("openblas")
+                    Conda.add("llvm-openmp", channel = "conda-forge")
+                    mkl_checked = true
                 catch
-                  mkl_checked = true
+                    mkl_checked = true
                 end   
-              end
-       #     
-              Conda.add("llvm-openmp", channel = "conda-forge")
-              PyCall.pyimport_conda("sklearn", "scikit-learn")
-
-            catch
-                @info("scikit-learn isn't properly installed."*
-                      "Please use PyCall default Conda or non-conda local python")
-                rethrow()
             end
+            Conda.add("llvm-openmp", channel = "conda-forge")
+            PyCall.pyimport_conda("sklearn", "scikit-learn")
+
+        catch
+            @info("scikit-learn isn't properly installed."*
+                    "Please use PyCall default Conda or non-conda local python")
+            rethrow()
+        end
 
     elseif Sys.islinux()
-        if Base.VERSION <= v"1.6.2" 
-            # GLIBCXX_3.4.26 
-            Conda.add("libstdcxx-ng>=3.4,<9.2", channel="conda-forge")
-        else 
-            # GLIBCXX_3.4.29 
-            # checked up to v1.8.0 
-            Conda.add("libstdcxx-ng>=3.4,<11.4", channel="conda-forge")
-        end 
+        if !libstdcxx_solved
+            version = _compatible_libstdcxx_ng_version()
+            Conda.add("conda", channel="anaconda")
+            Conda.add("libstdcxx-ng$version", channel="conda-forge")
+            if version == ">=3.4,<12.0" 
+                # https://github.com/scikit-learn/scikit-learn/pull/23990
+                Conda.add("scikit-learn<1.1", channel="conda-forge") 
+            end
+            libstdcxx_solved = true
+        end
+
         mod = PyCall.pyimport_conda("sklearn", "scikit-learn")
     else
         mod = PyCall.pyimport_conda("sklearn", "scikit-learn")

test/test_base.jl

@@ -59,7 +59,7 @@ function test_score_sample_weight()
     estimators = [DecisionTreeClassifier(max_depth=2),
                   DecisionTreeRegressor(max_depth=2)]
     sets = Any[datasets.load_iris(),
-               datasets.load_boston()]
+               datasets.fetch_california_housing()]
 
     for (est, ds) in zip(estimators, sets)
         fit!(est, ds["data"], ds["target"])