Fix ch.tensordot for numpy 1.14+ (mattloper#2)

.travis.yml

@@ -10,4 +10,6 @@ before_install:
 install:
   - pip install --upgrade pip
   - travis_wait pip install -r requirements.txt
-script: make test
+script:
+  - python -c 'import numpy; print numpy.version.version'
+  - make test

chumpy/ch_ops.py

@@ -793,13 +793,13 @@ def nonzero(a):
         a = a.r
     return np.nonzero(a)
 
-# Try to pull the code for tensordot in from numpy and reinterpret it using chumpy ops
-try:
-    import inspect
-    exec(''.join(inspect.getsourcelines(np.tensordot)[0]))
-    __all__ += ['tensordot']
-except:
-    pass
+# Pull the code for tensordot in from numpy and reinterpret it using chumpy ops
+import os
+source_path = os.path.join(os.path.dirname(os.path.abspath(__file__)), 'np_tensordot.py')
+with open(source_path, 'r') as f:
+    source_lines = f.readlines()
+exec(''.join(source_lines))
+__all__ += ['tensordot']
 
 
 

chumpy/np_tensordot.py

@@ -0,0 +1,228 @@
+# Up to numpy 1.13, the numpy implementation of tensordot could be
+# reinterpreted using chumpy. With numpy 1.14 the implementation started using
+# ufunc.multiply.reduce which can't be understood by chumpy. This is the
+# chumpy-compatible implementation of tensodrot from numpy 1.13.3.
+#
+# i.e.
+#
+# import inspect
+# with open('np_tensordot.py', 'w') as f:
+#     f.write(''.join(inspect.getsourcelines(np.tensordot)[0]))
+
+"""
+Copyright (c) 2005-2017, NumPy Developers.
+All rights reserved.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+
+    * Redistributions of source code must retain the above copyright
+       notice, this list of conditions and the following disclaimer.
+
+    * Redistributions in binary form must reproduce the above
+       copyright notice, this list of conditions and the following
+       disclaimer in the documentation and/or other materials provided
+       with the distribution.
+
+    * Neither the name of the NumPy Developers nor the names of any
+       contributors may be used to endorse or promote products derived
+       from this software without specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+"AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+"""
+
+def tensordot(a, b, axes=2):
+    """
+    Compute tensor dot product along specified axes for arrays >= 1-D.
+
+    Given two tensors (arrays of dimension greater than or equal to one),
+    `a` and `b`, and an array_like object containing two array_like
+    objects, ``(a_axes, b_axes)``, sum the products of `a`'s and `b`'s
+    elements (components) over the axes specified by ``a_axes`` and
+    ``b_axes``. The third argument can be a single non-negative
+    integer_like scalar, ``N``; if it is such, then the last ``N``
+    dimensions of `a` and the first ``N`` dimensions of `b` are summed
+    over.
+
+    Parameters
+    ----------
+    a, b : array_like, len(shape) >= 1
+        Tensors to "dot".
+
+    axes : int or (2,) array_like
+        * integer_like
+          If an int N, sum over the last N axes of `a` and the first N axes
+          of `b` in order. The sizes of the corresponding axes must match.
+        * (2,) array_like
+          Or, a list of axes to be summed over, first sequence applying to `a`,
+          second to `b`. Both elements array_like must be of the same length.
+
+    See Also
+    --------
+    dot, einsum
+
+    Notes
+    -----
+    Three common use cases are:
+        * ``axes = 0`` : tensor product :math:`a\\otimes b`
+        * ``axes = 1`` : tensor dot product :math:`a\\cdot b`
+        * ``axes = 2`` : (default) tensor double contraction :math:`a:b`
+
+    When `axes` is integer_like, the sequence for evaluation will be: first
+    the -Nth axis in `a` and 0th axis in `b`, and the -1th axis in `a` and
+    Nth axis in `b` last.
+
+    When there is more than one axis to sum over - and they are not the last
+    (first) axes of `a` (`b`) - the argument `axes` should consist of
+    two sequences of the same length, with the first axis to sum over given
+    first in both sequences, the second axis second, and so forth.
+
+    Examples
+    --------
+    A "traditional" example:
+
+    >>> a = np.arange(60.).reshape(3,4,5)
+    >>> b = np.arange(24.).reshape(4,3,2)
+    >>> c = np.tensordot(a,b, axes=([1,0],[0,1]))
+    >>> c.shape
+    (5, 2)
+    >>> c
+    array([[ 4400.,  4730.],
+           [ 4532.,  4874.],
+           [ 4664.,  5018.],
+           [ 4796.,  5162.],
+           [ 4928.,  5306.]])
+    >>> # A slower but equivalent way of computing the same...
+    >>> d = np.zeros((5,2))
+    >>> for i in range(5):
+    ...   for j in range(2):
+    ...     for k in range(3):
+    ...       for n in range(4):
+    ...         d[i,j] += a[k,n,i] * b[n,k,j]
+    >>> c == d
+    array([[ True,  True],
+           [ True,  True],
+           [ True,  True],
+           [ True,  True],
+           [ True,  True]], dtype=bool)
+
+    An extended example taking advantage of the overloading of + and \\*:
+
+    >>> a = np.array(range(1, 9))
+    >>> a.shape = (2, 2, 2)
+    >>> A = np.array(('a', 'b', 'c', 'd'), dtype=object)
+    >>> A.shape = (2, 2)
+    >>> a; A
+    array([[[1, 2],
+            [3, 4]],
+           [[5, 6],
+            [7, 8]]])
+    array([[a, b],
+           [c, d]], dtype=object)
+
+    >>> np.tensordot(a, A) # third argument default is 2 for double-contraction
+    array([abbcccdddd, aaaaabbbbbbcccccccdddddddd], dtype=object)
+
+    >>> np.tensordot(a, A, 1)
+    array([[[acc, bdd],
+            [aaacccc, bbbdddd]],
+           [[aaaaacccccc, bbbbbdddddd],
+            [aaaaaaacccccccc, bbbbbbbdddddddd]]], dtype=object)
+
+    >>> np.tensordot(a, A, 0) # tensor product (result too long to incl.)
+    array([[[[[a, b],
+              [c, d]],
+              ...
+
+    >>> np.tensordot(a, A, (0, 1))
+    array([[[abbbbb, cddddd],
+            [aabbbbbb, ccdddddd]],
+           [[aaabbbbbbb, cccddddddd],
+            [aaaabbbbbbbb, ccccdddddddd]]], dtype=object)
+
+    >>> np.tensordot(a, A, (2, 1))
+    array([[[abb, cdd],
+            [aaabbbb, cccdddd]],
+           [[aaaaabbbbbb, cccccdddddd],
+            [aaaaaaabbbbbbbb, cccccccdddddddd]]], dtype=object)
+
+    >>> np.tensordot(a, A, ((0, 1), (0, 1)))
+    array([abbbcccccddddddd, aabbbbccccccdddddddd], dtype=object)
+
+    >>> np.tensordot(a, A, ((2, 1), (1, 0)))
+    array([acccbbdddd, aaaaacccccccbbbbbbdddddddd], dtype=object)
+
+    """
+    try:
+        iter(axes)
+    except:
+        axes_a = list(range(-axes, 0))
+        axes_b = list(range(0, axes))
+    else:
+        axes_a, axes_b = axes
+    try:
+        na = len(axes_a)
+        axes_a = list(axes_a)
+    except TypeError:
+        axes_a = [axes_a]
+        na = 1
+    try:
+        nb = len(axes_b)
+        axes_b = list(axes_b)
+    except TypeError:
+        axes_b = [axes_b]
+        nb = 1
+
+    a, b = asarray(a), asarray(b)
+    as_ = a.shape
+    nda = a.ndim
+    bs = b.shape
+    ndb = b.ndim
+    equal = True
+    if na != nb:
+        equal = False
+    else:
+        for k in range(na):
+            if as_[axes_a[k]] != bs[axes_b[k]]:
+                equal = False
+                break
+            if axes_a[k] < 0:
+                axes_a[k] += nda
+            if axes_b[k] < 0:
+                axes_b[k] += ndb
+    if not equal:
+        raise ValueError("shape-mismatch for sum")
+
+    # Move the axes to sum over to the end of "a"
+    # and to the front of "b"
+    notin = [k for k in range(nda) if k not in axes_a]
+    newaxes_a = notin + axes_a
+    N2 = 1
+    for axis in axes_a:
+        N2 *= as_[axis]
+    newshape_a = (-1, N2)
+    olda = [as_[axis] for axis in notin]
+
+    notin = [k for k in range(ndb) if k not in axes_b]
+    newaxes_b = axes_b + notin
+    N2 = 1
+    for axis in axes_b:
+        N2 *= bs[axis]
+    newshape_b = (N2, -1)
+    oldb = [bs[axis] for axis in notin]
+
+    at = a.transpose(newaxes_a).reshape(newshape_a)
+    bt = b.transpose(newaxes_b).reshape(newshape_b)
+    res = dot(at, bt)
+    return res.reshape(olda + oldb)

circle.yml

@@ -15,4 +15,5 @@ dependencies:
 
 test:
   override:
+    - python -c 'import numpy; print numpy.version.version'
     - make test