scikit-learn · GaelVaroquaux · May 18, 2021 · May 5, 2021 · May 10, 2021 · May 10, 2021
diff --git a/.circleci/config.yml b/.circleci/config.yml
@@ -3,12 +3,12 @@ version: 2
 jobs:
   doc-min-dependencies:
     docker:
-      - image: circleci/python:3.7.3-stretch
+      - image: circleci/python:3.7.7-buster
     environment:
       - OMP_NUM_THREADS: 2
       - MKL_NUM_THREADS: 2
       - CONDA_ENV_NAME: testenv
-      - PYTHON_VERSION: 3.6
+      - PYTHON_VERSION: 3.7
       - NUMPY_VERSION: 'min'
       - SCIPY_VERSION: 'min'
       - MATPLOTLIB_VERSION: 'min'
@@ -47,7 +47,7 @@ jobs:
 
   doc:
     docker:
-      - image: circleci/python:3.7.3-stretch
+      - image: circleci/python:3.7.7-buster
     environment:
       - OMP_NUM_THREADS: 2
       - MKL_NUM_THREADS: 2
@@ -96,7 +96,7 @@ jobs:
 
   lint:
     docker:
-      - image: circleci/python:3.6
+      - image: circleci/python:3.7
     steps:
       - checkout
       - run: ./build_tools/circle/checkout_merge_commit.sh
@@ -130,7 +130,7 @@ jobs:
 
   deploy:
     docker:
-      - image: circleci/python:3.6
+      - image: circleci/python:3.7
     steps:
       - checkout
       - run: ./build_tools/circle/checkout_merge_commit.sh

diff --git a/.travis.yml b/.travis.yml
@@ -40,19 +40,11 @@ jobs:
         - CPU_COUNT=4
 
     # Linux environments to build the scikit-learn wheels for the ARM64
-    # architecture and Python 3.6 and newer. This is used both at release time
+    # architecture and Python 3.7 and newer. This is used both at release time
     # with the manual trigger in the commit message in the release branch and as
     # a scheduled task to build the weekly dev build on the main branch. The
     # weekly frequency is meant to avoid depleting the Travis CI credits too
     # fast.
-    - python: 3.6
-      os: linux
-      arch: arm64
-      if: type = cron or commit_message =~ /\[cd build\]/
-      env:
-        - BUILD_WHEEL=true
-        - CIBW_BUILD=cp36-manylinux_aarch64
-
     - python: 3.7
       os: linux
       arch: arm64

diff --git a/azure-pipelines.yml b/azure-pipelines.yml
@@ -11,7 +11,7 @@ jobs:
 - job: git_commit
   displayName: Get Git Commit
   pool:
-    vmImage: ubuntu-18.04
+    vmImage: ubuntu-20.04
   steps:
     - bash: |
         set -ex
@@ -38,7 +38,7 @@ jobs:
     )
   displayName: Linting
   pool:
-    vmImage: ubuntu-18.04
+    vmImage: ubuntu-20.04
   steps:
     - task: UsePythonVersion@0
       inputs:
@@ -57,7 +57,7 @@ jobs:
 - template: build_tools/azure/posix.yml
   parameters:
     name: Linux_Nightly
-    vmImage: ubuntu-18.04
+    vmImage: ubuntu-20.04
     dependsOn: [git_commit, linting]
     condition: |
       and(
@@ -83,7 +83,7 @@ jobs:
 - template: build_tools/azure/posix.yml
   parameters:
     name: Linux_Nightly_ICC
-    vmImage: ubuntu-18.04
+    vmImage: ubuntu-20.04
     dependsOn: [git_commit, linting]
     condition: |
       and(
@@ -105,7 +105,7 @@ jobs:
 - template: build_tools/azure/posix.yml
   parameters:
     name: Linux_Runs
-    vmImage: ubuntu-18.04
+    vmImage: ubuntu-20.04
     dependsOn: [git_commit]
     condition: |
       and(
@@ -119,10 +119,30 @@ jobs:
         BLAS: 'mkl'
         COVERAGE: 'true'
 
+# Check compilation with Ubuntu bionic 18.04 LTS and scipy from conda-forge
 - template: build_tools/azure/posix.yml
   parameters:
-    name: Linux
+    name: Ubuntu_Bionic
     vmImage: ubuntu-18.04
+    dependsOn: [git_commit, linting]
+    condition: |
+      and(
+        succeeded(),
+        not(contains(dependencies['git_commit']['outputs']['commit.message'], '[ci skip]')),
+        ne(variables['Build.Reason'], 'Schedule')
+      )
+    matrix:
+      py37_conda:
+        DISTRIB: 'conda'
+        PYTHON_VERSION: '3.7'
+        BLAS: 'openblas'
+        COVERAGE: 'false'
+        BUILD_WITH_ICC: 'false'
+
+- template: build_tools/azure/posix.yml
+  parameters:
+    name: Linux
+    vmImage: ubuntu-20.04
     dependsOn: [linting, git_commit]
     condition: |
       and(
@@ -132,32 +152,23 @@ jobs:
       )
     matrix:
       # Linux environment to test that scikit-learn can be built against
-      # versions of numpy, scipy with ATLAS that comes with Ubuntu Bionic 18.04
-      # i.e. numpy 1.13.3 and scipy 0.19
-      py36_ubuntu_atlas:
+      # versions of numpy, scipy with ATLAS that comes with Ubuntu Focal 20.04
+      # i.e. numpy 1.17.4 and scipy 1.3.3
+      ubuntu_atlas:
         DISTRIB: 'ubuntu'
-        PYTHON_VERSION: '3.6'
         JOBLIB_VERSION: 'min'
         PANDAS_VERSION: 'none'
         THREADPOOLCTL_VERSION: 'min'
-        PYTEST_VERSION: 'min'
-        PYTEST_XDIST_VERSION: 'none'
         COVERAGE: 'false'
-      # Linux + Python 3.6 build with OpenBLAS and without SITE_JOBLIB
-      py36_conda_openblas:
+      # Linux + Python 3.7 build with OpenBLAS and without SITE_JOBLIB
+      py37_conda_openblas:
         DISTRIB: 'conda'
-        PYTHON_VERSION: '3.6'
+        PYTHON_VERSION: '3.7'
         BLAS: 'openblas'
         NUMPY_VERSION: 'min'
         SCIPY_VERSION: 'min'
         MATPLOTLIB_VERSION: 'min'
-        # latest version of joblib available in conda for Python 3.6
-        JOBLIB_VERSION: '0.13.2'
         THREADPOOLCTL_VERSION: '2.0.0'
-        # temporary pin pytest due to unknown failure with pytest 5.4 and
-        # python 3.6
-        PYTEST_VERSION: 'min'
-        PYTEST_XDIST_VERSION: 'none'
       # Linux environment to test the latest available dependencies and MKL.
       # It runs tests requiring lightgbm, pandas and PyAMG.
       pylatest_pip_openblas_pandas:
@@ -171,7 +182,7 @@ jobs:
 - template: build_tools/azure/posix-32.yml
   parameters:
     name: Linux32
-    vmImage: ubuntu-18.04
+    vmImage: ubuntu-20.04
     dependsOn: [linting, git_commit]
     condition: |
       and(
@@ -180,14 +191,11 @@ jobs:
         ne(variables['Build.Reason'], 'Schedule')
       )
     matrix:
-      py36_ubuntu_atlas_32bit:
-        DISTRIB: 'ubuntu-32'
-        PYTHON_VERSION: '3.6'
+      debian_atlas_32bit:
+        DISTRIB: 'debian-32'
         JOBLIB_VERSION: 'min'
         # disable pytest xdist due to unknown bug with 32-bit container
         PYTEST_XDIST_VERSION: 'none'
-        # temporary pin pytest due to unknown failure with pytest 5.4 and
-        # python 3.6
         PYTEST_VERSION: 'min'
         THREADPOOLCTL_VERSION: 'min'
 
@@ -231,6 +239,6 @@ jobs:
         PYTHON_ARCH: '64'
         PYTEST_VERSION: '*'
         COVERAGE: 'true'
-      py36_pip_openblas_32bit:
-        PYTHON_VERSION: '3.6'
+      py37_pip_openblas_32bit:
+        PYTHON_VERSION: '3.7'
         PYTHON_ARCH: '32'
diff --git a/build_tools/azure/install.sh b/build_tools/azure/install.sh
@@ -70,9 +70,9 @@ elif [[ "$DISTRIB" == "ubuntu" ]]; then
     python -m pip install $(get_dep cython $CYTHON_VERSION) \
                           $(get_dep joblib $JOBLIB_VERSION)
 
-elif [[ "$DISTRIB" == "ubuntu-32" ]]; then
+elif [[ "$DISTRIB" == "debian-32" ]]; then
     apt-get update
-    apt-get install -y python3-dev python3-scipy python3-matplotlib libatlas3-base libatlas-base-dev python3-virtualenv python3-pandas ccache
+    apt-get install -y python3-dev python3-numpy python3-scipy python3-matplotlib libatlas3-base libatlas-base-dev python3-virtualenv python3-pandas ccache
 
     python3 -m virtualenv --system-site-packages --python=python3 $VIRTUALENV
     source $VIRTUALENV/bin/activate

@@ -45,7 +45,7 @@ jobs:
         -w /io
         --detach
         --name skcontainer
-        -e DISTRIB=ubuntu-32
+        -e DISTRIB=debian-32
         -e TEST_DIR=/temp_dir
         -e JUNITXML=$JUNITXML
         -e VIRTUALENV=testvenv
@@ -63,7 +63,7 @@ jobs:
         -e OMP_NUM_THREADS=$OMP_NUM_THREADS
         -e OPENBLAS_NUM_THREADS=$OPENBLAS_NUM_THREADS
         -e SKLEARN_SKIP_NETWORK_TESTS=$SKLEARN_SKIP_NETWORK_TESTS
-        i386/ubuntu:18.04
+        i386/debian:10.9
         sleep 1000000
       displayName: 'Start container'
     - script: >

diff --git a/build_tools/azure/test_script.sh b/build_tools/azure/test_script.sh
@@ -4,7 +4,7 @@ set -e
 
 if [[ "$DISTRIB" =~ ^conda.* ]]; then
     source activate $VIRTUALENV
-elif [[ "$DISTRIB" == "ubuntu" ]] || [[ "$DISTRIB" == "ubuntu-32" ]]; then
+elif [[ "$DISTRIB" == "ubuntu" ]] || [[ "$DISTRIB" == "debian-32" ]]; then
     source $VIRTUALENV/bin/activate
 fi
 

diff --git a/doc/conftest.py b/doc/conftest.py
@@ -7,6 +7,7 @@
 from sklearn.utils import IS_PYPY
 from sklearn.utils._testing import SkipTest
 from sklearn.utils._testing import check_skip_network
+from sklearn.utils.fixes import parse_version
 from sklearn.datasets import get_data_home
 from sklearn.datasets._base import _pkl_filepath
 from sklearn.datasets._twenty_newsgroups import CACHE_NAME
@@ -80,6 +81,10 @@ def setup_grid_search():
 def setup_preprocessing():
     try:
         import pandas  # noqa
+        if parse_version(pandas.__version__) < parse_version('1.1.0'):
+            raise SkipTest(
+                "Skipping preprocessing.rst, pandas version < 1.1.0"
+                )
     except ImportError:
         raise SkipTest("Skipping preprocessing.rst, pandas not installed")
 

diff --git a/doc/modules/sgd.rst b/doc/modules/sgd.rst
@@ -130,7 +130,7 @@ Using ``loss="log"`` or ``loss="modified_huber"`` enables the
 :math:`P(y|x)` per sample :math:`x`::
 
     >>> clf = SGDClassifier(loss="log", max_iter=5).fit(X, y)
-    >>> clf.predict_proba([[1., 1.]])
+    >>> clf.predict_proba([[1., 1.]]) # doctest: +SKIP
     array([[0.00..., 0.99...]])
 
 The concrete penalty can be set via the ``penalty`` parameter.

diff --git a/doc/tutorial/statistical_inference/supervised_learning.rst b/doc/tutorial/statistical_inference/supervised_learning.rst
@@ -173,7 +173,7 @@ Linear models: :math:`y = X\beta + \epsilon`
     >>> regr = linear_model.LinearRegression()
     >>> regr.fit(diabetes_X_train, diabetes_y_train)
     LinearRegression()
-    >>> print(regr.coef_)
+    >>> print(regr.coef_) # doctest: +SKIP
     [   0.30349955 -237.63931533  510.53060544  327.73698041 -814.13170937
       492.81458798  102.84845219  184.60648906  743.51961675   76.09517222]
 

diff --git a/doc/whats_new/v1.0.rst b/doc/whats_new/v1.0.rst
@@ -12,6 +12,12 @@ Version 1.0.0
 
 .. include:: changelog_legend.inc
 
+Minimal dependencies
+--------------------
+
+Version 1.0.0 of scikit-learn requires python 3.7+, numpy 1.14.5+ and
+scipy 1.1.0+. Optional minimal dependency is matplotlib 2.2.2+.
+
 Enforcing keyword-only arguments
 --------------------------------
 

diff --git a/pyproject.toml b/pyproject.toml
@@ -11,5 +11,5 @@ requires = [
     # see: https://github.com/scipy/oldest-supported-numpy/blob/master/setup.cfg
     "oldest-supported-numpy",
 
-    "scipy>=0.19.1",
+    "scipy>=1.1.0",
 ]
diff --git a/sklearn/_min_dependencies.py b/sklearn/_min_dependencies.py
@@ -5,12 +5,11 @@
 
 # numpy scipy and cython should by in sync with pyproject.toml
 if platform.python_implementation() == 'PyPy':
-    SCIPY_MIN_VERSION = '1.1.0'
     NUMPY_MIN_VERSION = '1.19.0'
 else:
-    SCIPY_MIN_VERSION = '0.19.1'
-    NUMPY_MIN_VERSION = '1.13.3'
+    NUMPY_MIN_VERSION = '1.14.5'
 
+SCIPY_MIN_VERSION = '1.1.0'
 JOBLIB_MIN_VERSION = '0.11'
 THREADPOOLCTL_MIN_VERSION = '2.0.0'
 PYTEST_MIN_VERSION = '5.0.1'
@@ -26,9 +25,9 @@
     'joblib': (JOBLIB_MIN_VERSION, 'install'),
     'threadpoolctl': (THREADPOOLCTL_MIN_VERSION, 'install'),
     'cython': (CYTHON_MIN_VERSION, 'build'),
-    'matplotlib': ('2.1.1', 'benchmark, docs, examples, tests'),
-    'scikit-image': ('0.13', 'docs, examples, tests'),
-    'pandas': ('0.25.0', 'benchmark, docs, examples, tests'),
+    'matplotlib': ('2.2.2', 'benchmark, docs, examples, tests'),
+    'scikit-image': ('0.14', 'docs, examples, tests'),
+    'pandas': ('0.23.4', 'benchmark, docs, examples, tests'),
     'seaborn': ('0.9.0', 'docs, examples'),
     'memory_profiler': ('0.57.0', 'benchmark, docs'),
     'pytest': (PYTEST_MIN_VERSION, 'tests'),

diff --git a/sklearn/decomposition/_truncated_svd.py b/sklearn/decomposition/_truncated_svd.py
@@ -87,18 +87,21 @@ class TruncatedSVD(TransformerMixin, BaseEstimator):
     Examples
     --------
     >>> from sklearn.decomposition import TruncatedSVD
-    >>> from scipy.sparse import random as sparse_random
-    >>> X = sparse_random(100, 100, density=0.01, format='csr',
-    ...                   random_state=42)
+    >>> from scipy.sparse import csr_matrix
+    >>> import numpy as np
+    >>> np.random.seed(0)
+    >>> X_dense = np.random.rand(100, 100)
+    >>> X_dense[:, 2 * np.arange(50)] = 0
+    >>> X = csr_matrix(X_dense)
     >>> svd = TruncatedSVD(n_components=5, n_iter=7, random_state=42)
     >>> svd.fit(X)
     TruncatedSVD(n_components=5, n_iter=7, random_state=42)
     >>> print(svd.explained_variance_ratio_)
-    [0.0646... 0.0633... 0.0639... 0.0535... 0.0406...]
+    [0.0157... 0.0512... 0.0499... 0.0479... 0.0453...]
     >>> print(svd.explained_variance_ratio_.sum())
-    0.286...
+    0.2102...
     >>> print(svd.singular_values_)
-    [1.553... 1.512...  1.510... 1.370... 1.199...]
+    [35.2410...  4.5981...   4.5420...  4.4486...  4.3288...]
 
     See Also
     --------

diff --git a/sklearn/ensemble/_hist_gradient_boosting/tests/test_loss.py b/sklearn/ensemble/_hist_gradient_boosting/tests/test_loss.py
@@ -98,7 +98,7 @@ def fprime2(x: np.ndarray) -> np.ndarray:
     optimum = optimum.ravel()
     assert_allclose(loss.inverse_link_function(optimum), y_true)
     assert_allclose(func(optimum), 0, atol=1e-14)
-    assert_allclose(get_gradients(y_true, optimum), 0, atol=1e-7)
+    assert_allclose(get_gradients(y_true, optimum), 0, atol=1e-6)
 
 
 @pytest.mark.parametrize('loss, n_classes, prediction_dim', [