8000 ENH: `__array_function__` for `np.einsum` and `np.block` by shoyer · Pull Request #12163 · numpy/numpy · GitHub
[go: up one dir, main page]
Skip to content

ENH: __array_function__ for np.einsum and np.block #12163

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 6 commits into from
Oct 26, 2018
Merged

ENH: __array_function__ for np.einsum and np.block #12163

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
dfab760
ENH: __array_function__ for np.einsum and np.block
shoyer Oct 13, 2018
024c728
MAINT: add comment explaining _block_dispatcher
shoyer Oct 19, 2018
4701c31
TST: tests for _block_dispatcher
shoyer Oct 25, 2018
ea7ff5f
Merge branch 'einsum-dispatch' of github.com:shoyer/numpy into einsum…
shoyer Oct 25, 2018
2dfb68e
Merge branch 'master' into einsum-dispatch
shoyer Oct 25, 2018
b9a13b5
MAINT: fix block dispatcher test
shoyer Oct 25, 2018
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
Next Next commit
ENH: __array_function__ for np.einsum and np.block
  • Loading branch information
@shoyer
shoyer committed Oct 13, 2018
commit dfab760b4a328d9fa29cef123e0fe8e2926b0c8c
21 changes: 21 additions & 0 deletions numpy/core/einsumfunc.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -9,6 +9,7 @@
from numpy.compat import basestring
from numpy.core.multiarray import c_einsum
from numpy.core.numeric import asanyarray, tensordot
from numpy.core.overrides import array_function_dispatch

__all__ = ['einsum', 'einsum_path']

Expand Down Expand Up @@ -689,6 +690,17 @@ def _parse_einsum_input(operands):
return (input_subscripts, output_subscript, operands)


def _einsum_path_dispatcher(*operands, **kwargs):
# NOTE: technically, we should only dispatch on array-like arguments, not
# subscripts (given as strings). But separating operands into
# arrays/subscripts is a little tricky/slow (given einsum's two supported
# signatures), so as a practical shortcut we dispatch on everything.
# Strings will be ignored for dispatching since they don't define
# __array_function__.
return operands


@array_function_dispatch(_einsum_path_dispatcher)
def einsum_path(*operands, **kwargs):
"""
einsum_path(subscripts, *operands, optimize='greedy')
Expand Down Expand Up @@ -980,7 +992,16 @@ def einsum_path(*operands, **kwargs):
return (path, path_print)


def _einsum_dispatcher(*operands, **kwargs):
# Arguably we dispatch on more arguments that we really should; see note in
# _einsum_path_dispatcher for why.
for op in operands:
yield op
yield kwargs.get('out')


# Rewrite einsum to handle different cases
@array_function_dispatch(_einsum_dispatcher)
def einsum(*operands, **kwargs):
"""
einsum(subscripts, *operands, out=None, dtype=None, order='K',
Expand Down
11 changes: 10 additions & 1 deletion numpy/core/shape_base.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -499,7 +499,16 @@ def _block(arrays, max_depth, result_ndim, depth=0):
return _atleast_nd(arrays, result_ndim)


# TODO: support array_function_dispatch
def _block_dispatcher(arrays):
if type(arrays) is list:
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I'm not sure if we'll be breaking backward compatibility by doing this. Maybe isinstance(arrays, collections.Iterable) and not hasattr(arrays, '__array_function__')?

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

My reasoning for picking type(arrays) is list is that it matches np.block(), which hard codes list (not even tuple). And we know that list.__array_function__ will never exist.

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Okay, in that case ignore my comment. I was worried about the case where we might be inadvertently supporting other kinds of containers in the function but not the dispatcher. A comment would be nice, in any case.

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Yes, a comment is definitely a good idea here. Done!

for subarrays in arrays:
for subarray in _block_dispatcher(subarrays):
yield subarray
else:
yield arrays


@array_function_dispatch(_block_dispatcher)
def block(arrays):
"""
Assemble an nd-array from nested lists of blocks.
Expand Down
0