Merge branch 'master' into python3.13

.github/workflows/check.yml

@@ -22,12 +22,12 @@ jobs:
         #- ruff
         - package
     steps:
-    - uses: actions/checkout@v3
-    - uses: actions/setup-python@v4
+    - uses: actions/checkout@v4
+    - uses: actions/setup-python@v5.1.1
       with:
         python-version: '3.10'
     - name: Install build tools
       run: |
         python -m pip install tox wheel
     - name: Run ${{ matrix.env }}
-      run: tox -e ${{ matrix.env }}
+      run: python -m tox -e ${{ matrix.env }}

.github/workflows/publish.yml

@@ -13,8 +13,8 @@ jobs:
   publish:
     runs-on: ubuntu-latest
     steps:
-    - uses: actions/checkout@v3
-    - uses: actions/setup-python@v4
+    - uses: actions/checkout@v4
+    - uses: actions/setup-python@v5.1.1
       with:
         python-version: '3.10'
     - name: Install build tools

.github/workflows/test.yml

@@ -23,29 +23,28 @@ jobs:
         - macos-latest
         - windows-latest
         python-version:
-        - '3.7'
-        - '3.8'
         - '3.9'
         - '3.10'
         - '3.11'
-        - pypy-3.7
-        - pypy-3.8
+        - '3.12'
         - pypy-3.9
         - pypy-3.10
+        # TODO: bring this back later
         exclude:
-        - { platform: windows-latest, python-version: pypy-3.7 }
+        - platform: windows-latest
+          python-version: 'pypy-3.10'
     steps:
-    - uses: actions/checkout@v3
-    - uses: actions/setup-python@v4
+    - uses: actions/checkout@v4
+    - uses: actions/setup-python@v5.1.1
       with:
         python-version: ${{ matrix.python-version }}
     - name: Install Python build tools
       run: python -m pip install tox wheel
     - name: Run tests
-      run: tox run -e py
+      run: python -m tox run -e py
     - name: Install Scipy prerequisites for Ubuntu
       if: startsWith(matrix.platform, 'ubuntu')
       run: sudo apt-get install libopenblas-dev
     - name: Run tests with scipy
       if: startsWith(matrix.platform, 'ubuntu') || startsWith(matrix.python-version, 'pypy') != true
-      run: tox run -e py-scipy
+      run: python -m tox run -e py-scipy

README.md

@@ -1,8 +1,8 @@
-__Note: Autograd is still being maintained but is no longer actively developed.
-The main developers (Dougal Maclaurin, David Duvenaud, Matt Johnson, and Jamie
-Townsend) are now working on [JAX](https://github.com/google/jax), with Dougal and Matt
-working on it full-time. JAX combines a new version of Autograd with extra
-features such as jit compilation.__
+__Note (updated May 2024): Autograd is not currently maintained, and the
+authors do not plan to respond to future issues or pull requests. Those looking
+for a similar user experience (with powerful additional features) can consider
+using [JAX](https://github.com/google/jax), a successor to Autograd designed by
+the same authors.__
 
 # Autograd  [![Checks status][checks-badge]][checks-url] [![Tests status][tests-badge]][tests-url] [![Publish status][publish-badge]][publish-url] [![asv][asv-badge]](#)
 

autograd/builtins.py

@@ -1,4 +1,3 @@
-from six import with_metaclass
 from .util import subvals
 from .extend import (Box, primitive, notrace_primitive, VSpace, vspace,
                      SparseObject, defvjp, defvjp_argnum, defjvp, defjvp_argnum)
@@ -86,24 +85,25 @@ def fwd_grad_make_sequence(argnum, g, ans, seq_type, *args, **kwargs):
 defjvp_argnum(make_sequence, fwd_grad_make_sequence)
 
 
-class TupleMeta(type_):
+class TupleMeta(type(tuple_)):
     def __instancecheck__(self, instance):
         return isinstance(instance, tuple_)
-class tuple(with_metaclass(TupleMeta, tuple_)):
+
+class tuple(tuple_, metaclass=TupleMeta):
     def __new__(cls, xs):
         return make_sequence(tuple_, *xs)
 
 class ListMeta(type_):
     def __instancecheck__(self, instance):
         return isinstance(instance, list_)
-class list(with_metaclass(ListMeta, list_)):
+class list(list_, metaclass=ListMeta):
     def __new__(cls, xs):
-        return make_sequence(list_,  *xs)
+        return make_sequence(list_, *xs)
 
 class DictMeta(type_):
     def __instancecheck__(self, instance):
         return isinstance(instance, dict_)
-class dict(with_metaclass(DictMeta, dict_)):
+class dict(dict_, metaclass=DictMeta):
     def __new__(cls, *args, **kwargs):
         result = dict_(*args, **kwargs)
         if result:

autograd/numpy/numpy_boxes.py

@@ -66,7 +66,13 @@ def __hash__(self): return id(self)
 # Flatten has no function, only a method.
 setattr(ArrayBox, 'flatten', anp.__dict__['ravel'])
 
-if np.__version__ >= '1.25':
+if np.lib.NumpyVersion(np.__version__) >= '2.0.0':
+    SequenceBox.register(np.linalg._linalg.EigResult)
+    SequenceBox.register(np.linalg._linalg.EighResult)
+    SequenceBox.register(np.linalg._linalg.QRResult)
+    SequenceBox.register(np.linalg._linalg.SlogdetResult)
+    SequenceBox.register(np.linalg._linalg.SVDResult)
+elif np.__version__ >= '1.25':
     SequenceBox.register(np.linalg.linalg.EigResult)
     SequenceBox.register(np.linalg.linalg.EighResult)
     SequenceBox.register(np.linalg.linalg.QRResult)

autograd/numpy/numpy_jvps.py

@@ -1,3 +1,4 @@
+import numpy as onp
 from . import numpy_wrapper as anp
 from .numpy_vjps import (untake, balanced_eq, match_complex, replace_zero,
                          dot_adjoint_0, dot_adjoint_1, tensordot_adjoint_0,
@@ -210,7 +211,8 @@ def fwd_grad_sort(g, ans, x, axis=-1, kind='quicksort', order=None):
     sort_perm = anp.argsort(x, axis, kind, order)
     return g[sort_perm]
 defjvp(anp.sort, fwd_grad_sort)
-defjvp(anp.msort, lambda g, ans, x: fwd_grad_sort(g, ans, x, axis=0))
+if onp.lib.NumpyVersion(onp.__version__) < '2.0.0':
+    defjvp(anp.msort, lambda g, ans, x: fwd_grad_sort(g, ans, x, axis=0))
 
 def fwd_grad_partition(g, ans, x, kth, axis=-1, kind='introselect', order=None):
     partition_perm = anp.argpartition(x, kth, axis, kind, order)

autograd/numpy/numpy_vjps.py

@@ -1,5 +1,4 @@
 from __future__ import absolute_import
-from six import string_types
 from functools import partial
 import numpy as onp
 from ..util import func
@@ -560,7 +559,8 @@ def grad_sort(ans, x, axis=-1, kind='quicksort', order=None):
     sort_perm = anp.argsort(x, axis, kind, order)
     return lambda g: unpermuter(g, sort_perm)
 defvjp(anp.sort, grad_sort)
-defvjp(anp.msort, grad_sort)  # Until multi-D is allowed, these are the same.
+if onp.lib.NumpyVersion(onp.__version__) < '2.0.0':
+    defvjp(anp.msort, grad_sort)  # Until multi-D is allowed, these are the same.
 
 def grad_partition(ans, x, kth, axis=-1, kind='introselect', order=None):
     #TODO: Cast input with np.asanyarray()
@@ -588,7 +588,7 @@ def grad_einsum(argnum, ans, operands_, kwargs):
     result_meta = anp.metadata(operands_[argnum])
     def vjp(g):
         operands = operands_
-        if isinstance(operands[0], string_types):  # using "ijk" convention.
+        if isinstance(operands[0], str):  # using "ijk" convention.
             in_subs, out_subs, _ = anp.parse_einsum_input(*operands)
             string, operands = operands[0], operands[1:]
 

autograd/numpy/numpy_vspaces.py

@@ -4,7 +4,7 @@
 
 class ArrayVSpace(VSpace):
     def __init__(self, value):
-        value = np.array(value, copy=False)
+        value = np.asarray(value)
         self.shape = value.shape
         self.dtype = value.dtype
 
@@ -66,7 +66,19 @@ def _covector(self, x):
     ComplexArrayVSpace.register(type_)
 
 
-if np.__version__ >= '1.25':
+if np.lib.NumpyVersion(np.__version__) >= '2.0.0':
+    class EigResultVSpace(NamedTupleVSpace):     seq_type = np.linalg._linalg.EigResult
+    class EighResultVSpace(NamedTupleVSpace):    seq_type = np.linalg._linalg.EighResult
+    class QRResultVSpace(NamedTupleVSpace):      seq_type = np.linalg._linalg.QRResult
+    class SlogdetResultVSpace(NamedTupleVSpace): seq_type = np.linalg._linalg.SlogdetResult
+    class SVDResultVSpace(NamedTupleVSpace):     seq_type = np.linalg._linalg.SVDResult
+
+    EigResultVSpace.register(np.linalg._linalg.EigResult)
+    EighResultVSpace.register(np.linalg._linalg.EighResult)
+    QRResultVSpace.register(np.linalg._linalg.QRResult)
+    SlogdetResultVSpace.register(np.linalg._linalg.SlogdetResult)
+    SVDResultVSpace.register(np.linalg._linalg.SVDResult)
+elif np.__version__ >= '1.25':
     class EigResultVSpace(NamedTupleVSpace):     seq_type = np.linalg.linalg.EigResult
     class EighResultVSpace(NamedTupleVSpace):    seq_type = np.linalg.linalg.EighResult
     class QRResultVSpace(NamedTupleVSpace):      seq_type = np.linalg.linalg.QRResult

autograd/scipy/signal.py

@@ -6,7 +6,7 @@
 from autograd.extend import primitive, defvjp
 
 from numpy.lib.stride_tricks import as_strided
-from six import iteritems
+
 
 @primitive
 def convolve(A, B, axes=None, dot_axes=[(),()], mode='full'):
@@ -79,8 +79,8 @@ def parse_axes(A_shape, B_shape, conv_axes, dot_axes, mode):
                    'conv'     : tuple(range(i2, i3))}
     conv_shape = (compute_conv_size(A_shape[i], B_shape[j], mode)
                   for i, j in zip(axes['A']['conv'], axes['B']['conv']))
-    shapes = {'A'   : {s : (A_shape[i] for i in ax) for s, ax in iteritems(axes['A'])},
-              'B'   : {s : (B_shape[i] for i in ax) for s, ax in iteritems(axes['B'])}}
+    shapes = {'A': {s: tuple(A_shape[i] for i in ax) for s, ax in axes['A'].items()},
+              'B': {s: tuple(B_shape[i] for i in ax) for s, ax in axes['B'].items()}}
     shapes['out'] = {'ignore_A' : shapes['A']['ignore'],
                      'ignore_B' : shapes['B']['ignore'],
                      'conv'     : conv_shape}

conda_recipe/conda.yaml

@@ -14,7 +14,8 @@ source:
 requirements:
   build:
     - python
-    - setuptools
+    - hatch
+    - hatchling
     - future
     - numpy >=1.9
 
@@ -24,7 +25,7 @@ requirements:
     - numpy >=1.9
 
 build:
-  script: python setup.py build && python setup.py install
+  script: pip install . --no-deps
 
 test:
   # Python imports

examples/data_mnist.py

@@ -1,9 +1,6 @@
 from __future__ import absolute_import
 from __future__ import print_function
 import sys
-if sys.version < "3":
-    from future.standard_library import install_aliases
-    install_aliases()
 
 import os
 import gzip

pyproject.toml

@@ -1,11 +1,12 @@
 [build-system]
-requires = ["setuptools>=44"]
-build-backend = "setuptools.build_meta"
+requires = ["hatchling"]
+build-backend = "hatchling.build"
 
 [project]
 name = "autograd"
 version = "1.6.3"
-description = "Efficiently computes derivatives of numpy code."
+requires-python = ">=3.8"
+description = "Efficiently computes derivatives of NumPy code."
 readme = "README.md"
 license = {file = "license.txt"}
 authors = [
@@ -16,22 +17,20 @@ authors = [
 ]
 maintainers = [
   {name = "Jamie Townsend", email = "j.h.n.townsend@uva.nl"},
+  {name = "Fabian Joswig", email = "fabian.joswig@uni-muenster.de"},
+  {name = "Agriya Khetarpal", email = "agriyakhetarpal@outlook.com"},
 ]
 classifiers = [
   "Development Status :: 4 - Beta",
   "Intended Audience :: Information Technology",
   "Intended Audience :: Science/Research",
   "License :: OSI Approved :: MIT License",
-  "Programming Language :: Python :: 2",
-  "Programming Language :: Python :: 2.7",
-  "Programming Language :: Python :: 3",
-  "Programming Language :: Python :: 3.5",
-  "Programming Language :: Python :: 3.6",
-  "Programming Language :: Python :: 3.7",
   "Programming Language :: Python :: 3.8",
   "Programming Language :: Python :: 3.9",
   "Programming Language :: Python :: 3.10",
   "Programming Language :: Python :: 3.11",
+  "Programming Language :: Python :: 3.12",
+  "Topic :: Scientific/Engineering",
 ]
 keywords = [
   "Automatic differentiation",
@@ -41,14 +40,16 @@ keywords = [
   "optimization",
   "neural networks",
   "Python",
-  "Numpy",
-  "Scipy",
+  "NumPy",
+  "SciPy",
 ]
 dependencies = [
-  "numpy>=1.12",
-  "six",
-  "future>=0.15.2; python_version < '3'",
+  "numpy",
 ]
+# dynamic = ["version"]
+
+[project.urls]
+Source = "https://github.com/HIPS/autograd"
 
 [project.optional-dependencies]
 scipy = [
@@ -69,12 +70,3 @@ extend-exclude = []
 extend-ignore = ["E731"]
 extend-select = ["I", "W"]
 line-length = 109
-
-[tool.setuptools]
-packages=[
-  "autograd",
-  "autograd.numpy",
-  "autograd.scipy",
-  "autograd.scipy.stats",
-  "autograd.misc",
-]

tests/test_linalg.py

@@ -1,6 +1,7 @@
 from __future__ import absolute_import
 from builtins import range
 import itertools
+import numpy as onp
 import autograd.numpy as np
 import autograd.numpy.random as npr
 from autograd.test_util import check_grads
@@ -94,7 +95,10 @@ def test_solve_arg1_3d():
     D = 4
     A = npr.randn(D+1, D, D) + 5*np.eye(D)
     B = npr.randn(D+1, D)
-    fun = lambda A: np.linalg.solve(A, B)
+    if onp.lib.NumpyVersion(onp.__version__) < '2.0.0':
+        fun = lambda A: np.linalg.solve(A, B)
+    else:
+        fun = lambda A: np.linalg.solve(A, B[..., None])[..., 0]
     check_grads(fun)(A)
 
 def test_solve_arg1_3d_3d():

tests/test_systematic.py

@@ -1,4 +1,5 @@
 from __future__ import absolute_import
+import numpy as onp
 import autograd.numpy.random as npr
 import autograd.numpy as np
 import operator as op
@@ -43,7 +44,7 @@ def test_log1p():   unary_ufunc_check(np.log1p, lims=[0.2, 2.0])
 def test_log2():    unary_ufunc_check(np.log2,  lims=[0.2, 2.0])
 def test_rad2deg(): unary_ufunc_check(lambda x : np.rad2deg(x)/50.0, test_complex=False)
 def test_radians(): unary_ufunc_check(np.radians, test_complex=False)
-def test_sign():    unary_ufunc_check(np.sign)
+def test_sign():    unary_ufunc_check(np.sign, test_complex=False)
 def test_sin():     unary_ufunc_check(np.sin)
 def test_sinh():    unary_ufunc_check(np.sinh)
 def test_sqrt():    unary_ufunc_check(np.sqrt, lims=[1.0, 3.0])
@@ -155,7 +156,8 @@ def test_fmin(): combo_check(np.fmin, [0, 1])(
                             [R(1), R(1,4), R(3, 4)])
 
 def test_sort():       combo_check(np.sort, [0])([R(1), R(7)])
-def test_msort():     combo_check(np.msort, [0])([R(1), R(7)])
+if onp.lib.NumpyVersion(onp.__version__) < '2.0.0':
+    def test_msort():     combo_check(np.msort, [0])([R(1), R(7)])
 def test_partition(): combo_check(np.partition, [0])(
                                   [R(7), R(14)], kth=[0, 3, 6])
 

tox.ini

@@ -1,6 +1,8 @@
 # Tox (https://tox.wiki/) - run tests in isolation using virtualenv.
 # Also contains config settings for tools that don't look into pyproject.toml.
 
+# TODO: Migrate to tool.hatch.run or noxfile.py
+
 [tox]
 envlist =
     ruff
@@ -33,10 +35,10 @@ deps = pyclean
 commands = pyclean {posargs:. --debris --erase junit-report.xml --yes}
 
 [testenv:ruff]
-description = Lightening-fast linting for Python
+description = Lightning-fast linting for Python
 skip_install = true
 deps = ruff
-commands = ruff {posargs:.}
+commands = ruff check {posargs:.}  # TODO: Fix style failures
 
 [testenv:package]
 description = Build package and check metadata (or upload package)