diff options
| -rw-r--r-- | doc/neps/nep-0018-array-function-protocol.rst | 265 |
1 files changed, 169 insertions, 96 deletions
diff --git a/doc/neps/nep-0018-array-function-protocol.rst b/doc/neps/nep-0018-array-function-protocol.rst index 943ca4cbf..d8aaf7250 100644 --- a/doc/neps/nep-0018-array-function-protocol.rst +++ b/doc/neps/nep-0018-array-function-protocol.rst @@ -1,6 +1,6 @@ -================================================== -NEP: Dispatch Mechanism for NumPy's high level API -================================================== +============================================= +Dispatch Mechanism for NumPy's high level API +============================================= :Author: Stephan Hoyer <shoyer@google.com> :Author: Matthew Rocklin <mrocklin@gmail.com> @@ -11,11 +11,11 @@ NEP: Dispatch Mechanism for NumPy's high level API Abstact ------- -We propose a protocol to allow arguments of numpy functions to define -how that function operates on them. This allows other libraries that -implement NumPy's high level API to reuse Numpy functions. This allows -libraries that extend NumPy's high level API to apply to more NumPy-like -libraries. +We propose the ``__array_function__`` protocol, to allow arguments of numpy +functions to define how that function operates on them. This will allow +using NumPy as a high level API for efficient multi-dimensional array +operations, even with array implementations that differ greatly from +``numpy.ndarray``. Detailed description -------------------- @@ -71,10 +71,13 @@ We propose adding support for a new protocol in NumPy, ``__array_function__``. This protocol is intended to be a catch-all for NumPy functionality that -is not covered by existing protocols, like reductions (like ``np.sum``) -or universal functions (like ``np.exp``). The semantics are very similar -to ``__array_ufunc__``, except the operation is specified by an -arbitrary callable object rather than a ufunc instance and method. +is not covered by the ``__array_func__`` protocol for universal functions +(like ``np.exp``). The semantics are very similar to ``__array_ufunc__``, except +the operation is specified by an arbitrary callable object rather than a ufunc +instance and method. + +A prototype implementation with microbenchmark results can be found in +`this notebook <https://nbviewer.jupyter.org/gist/shoyer/1f0a308a06cd96df20879a1ddb8f0006>`_. The interface ~~~~~~~~~~~~~ @@ -84,27 +87,28 @@ We propose the following signature for implementations of .. code-block:: python - def __array_function__(self, func, types, args, kwargs) + def __array_function__(self, func, possibly_overloaded, args, kwargs) - ``func`` is an arbitrary callable exposed by NumPy's public API, which was called in the form ``func(*args, **kwargs)``. -- ``types`` is a list of types for all arguments to the original NumPy - function call that will be checked for an ``__array_function__`` - implementation. -- The tuple ``args`` and dict ``**kwargs`` are directly passed on from the +- ``types`` is a list of argument types from the original NumPy + function call that implement ``__array_function__``, in the order in which + they will be called. +- The tuple ``args`` and dict ``kwargs`` are directly passed on from the original call. Unlike ``__array_ufunc__``, there are no high-level guarantees about the type of ``func``, or about which of ``args`` and ``kwargs`` may contain objects implementing the array API. As a convenience for ``__array_function__`` -implementors of the NumPy API, the ``types`` keyword contains a list of all -types that implement the ``__array_function__`` protocol. This allows -downstream implementations to quickly determine if they are likely able to -support the operation. +implementors, ``types`` contains a list of argument types with an +``'__array_function__'`` attribute. This allows downstream implementations to +quickly determine if they are likely able to support the operation. Still be determined: what guarantees can we offer for ``types``? Should we promise that types are unique, and appear in the order in which they -are checked? +are checked? Should we pass in arguments directly instead, either the full +list of arguments in ``relevant_arguments`` (see below) or a single argument +for each unique type? Example for a project implementing the NumPy API ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ @@ -145,7 +149,9 @@ This will require two changes within the Numpy codebase: 1. A function to inspect available inputs, look for the ``__array_function__`` attribute on those inputs, and call those methods appropriately until one succeeds. This needs to be fast in the - common all-NumPy case. + common all-NumPy case, and have acceptable performance (no worse than + linear time) even if the number of overloaded inputs is large (e.g., + as might be the case for `np.concatenate`). This is one additional function of moderate complexity. 2. Calling this function within all relevant Numpy functions. @@ -187,15 +193,15 @@ of these may decide that, given the available inputs, they are unable to determine the correct result. How do we call the right one? If several are valid then which has precedence? -The rules for dispatch with ``__array_function__`` match those for -``__array_ufunc__`` (see +For the most part, the rules for dispatch with ``__array_function__`` +match those for ``__array_ufunc__`` (see `NEP-13 <http://www.numpy.org/neps/nep-0013-ufunc-overrides.html>`_). In particular: - NumPy will gather implementations of ``__array_function__`` from all specified inputs and call them in order: subclasses before - superclasses, and otherwise left to right. Note that in some edge cases, - this differs slightly from the + superclasses, and otherwise left to right. Note that in some edge cases + involving subclasses, this differs slightly from the `current behavior <https://bugs.python.org/issue30140>`_ of Python. - Implementations of ``__array_function__`` indicate that they can handle the operation by returning any value other than @@ -203,11 +209,44 @@ In particular: - If all ``__array_function__`` methods return ``NotImplemented``, NumPy will raise ``TypeError``. +One deviation from the current behavior of ``__array_ufunc__`` is that NumPy +will only call ``__array_function__`` on the *first* argument of each unique +type. This matches Python's +`rule for calling reflected methods <https://docs.python.org/3/reference/datamodel.html#object.__ror__>`_, +and this ensures that checking overloads has acceptable performance even when +there are a large number of overloaded arguments. + +Special handling of ``numpy.ndarray`` +''''''''''''''''''''''''''''''''''''' + +The use cases for subclasses with ``__array_function__`` are the same as those +with ``__array_ufunc__``, so ``numpy.ndarray`` should also define a +``__array_function__`` method mirroring ``ndarray.__array_ufunc__``: + +.. code:: python + + def __array_function__(self, func, types, args, kwargs): + # Cannot handle items that have __array_function__ other than our own. + for t in types: + if (hasattr(t, '__array_function__') and + t.__array_function__ is not ndarray.__array_function__): + return NotImplemented + + # Arguments contain no overrides, so we can safely call the + # overloaded function again. + return func(*args, **kwargs) + +To avoid recursion, the dispatch rules for ``__array_function__`` need also +the same special case they have for ``__array_ufunc__``: any arguments with an +``__array_function__`` method that is identical to +``numpy.ndarray.__array_function__`` are not be called as +``__array_function__`` implementations. + Changes within Numpy functions ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ -Given a function defined above, for now call it -``do_array_function_dance``, we now need to call that function from +Given a function defining the above behavior, for now call it +``try_array_function_override``, we now need to call that function from within every relevant Numpy function. This is a pervasive change, but of fairly simple and innocuous code that should complete quickly and without effect if no arguments implement the ``__array_function__`` @@ -216,8 +255,10 @@ might be affected by this change: .. code:: python + import itertools + def broadcast_to(array, shape, subok=False): - success, value = do_array_function_dance( + success, value = try_array_function_override( func=broadcast_to, relevant_arguments=[array], args=(array,), @@ -225,41 +266,51 @@ might be affected by this change: if success: return value - ... # continue with the definition of broadcast_to + ... # continue with the definition of broadcast_to def concatenate(arrays, axis=0, out=None) - success, value = do_array_function_dance( + success, value = try_array_function_override( func=concatenate, - relevant_arguments=[arrays, out], + relevant_arguments=itertools.chain(arrays, [out]), args=(arrays,), kwargs=dict(axis=axis, out=out)) if success: return value - ... # continue with the definition of concatenate + ... # continue with the definition of concatenate The list of objects passed to ``relevant_arguments`` are those that should be inspected for ``__array_function__`` implementations. -Alternatively, we could write these overloads with a decorator, e.g., +Our microbenchmark results show that a pure Python implementation of +``try_array_function_override`` adds approximately 2-4 microseconds of +overhead to each function call without any overloaded arguments. +This is acceptable for functions implemented in Python but probably too +slow for functions written in C. Fortunately, we expect significantly less +overhead with a C implementation of ``try_array_function_override``. + +A more succinct alternative would be to write these overloads with a decorator, +e.g., .. code:: python @overload_for_array_function(['array']) def broadcast_to(array, shape, subok=False): - ... # continue with the definition of broadcast_to + ... # continue with the definition of broadcast_to @overload_for_array_function(['arrays', 'out']) def concatenate(arrays, axis=0, out=None): - ... # continue with the definition of concatenate + ... # continue with the definition of concatenate The decorator ``overload_for_array_function`` would be written in terms -of ``do_array_function_dance``. +of ``try_array_function_override``. -The downside of this approach would be a loss of introspection capability -for NumPy functions on Python 2, since this requires the use of -``inspect.Signature`` (only available on Python 3). However, NumPy won't -be supporting Python 2 for `very much longer <http://www.numpy.org/neps/nep-0014-dropping-python2.7-proposal.html>`_. +Implementing it cleanly would require the signature preserving version of +``functools.wraps`` (only available on Python 3) and the standard libray's +``inspect`` module. NumPy won't be supporting Python 2 for +`very much longer <http://www.numpy.org/neps/nep-0014-dropping-python2.7-proposal.html>`_, but a bigger issue is +performance: ``inspect`` is written in pure Python, so our prototype +decorator pretty slow, adding about 15 microseonds of overhead. Use outside of NumPy ~~~~~~~~~~~~~~~~~~~~ @@ -276,8 +327,8 @@ to be explicitly recognized. Libraries like Dask, CuPy, and Autograd already wrap a limited subset of SciPy functionality (e.g., ``scipy.linalg``) similarly to how they wrap NumPy. -If we want to do this, we should consider exposing the helper function -``do_array_function_dance()`` above as a public API. +If we want to do this, we should expose the helper function +``try_array_function_override()`` as a public API. Non-goals --------- @@ -358,6 +409,11 @@ functions from ``numpy`` itself are already overloaded (but inadequately), so confusion about high vs. low level APIs in NumPy would still persist. +Alternatively, a separate namespace, e.g., ``numpy.array_only``, could be +created for a non-overloaded version of NumPy's high level API, for cases +where performance with NumPy arrays is a critical concern. This has most +of the same downsides as the separate namespace. + Multiple dispatch ~~~~~~~~~~~~~~~~~ @@ -385,30 +441,33 @@ Implementations in terms of a limited core API ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ The internal implemenations of some NumPy functions is extremely simple. -For example: - ``np.stack()`` is implemented in only a few lines of code -by combining indexing with ``np.newaxis``, ``np.concatenate`` and the -``shape`` attribute. - ``np.mean()`` is implemented internally in terms -of ``np.sum()``, ``np.divide()``, ``.astype()`` and ``.shape``. +For example: + +- ``np.stack()`` is implemented in only a few lines of code by combining + indexing with ``np.newaxis``, ``np.concatenate`` and the ``shape`` attribute. +- ``np.mean()`` is implemented internally in terms of ``np.sum()``, + ``np.divide()``, ``.astype()`` and ``.shape``. This suggests the possibility of defining a minimal "core" ndarray interface, and relying upon it internally in NumPy to implement the full API. This is an attractive option, because it could significantly reduce the work required for new array implementations. -However, this also comes with several downsides: 1. The details of how -NumPy implements a high-level function in terms of overloaded functions -now becomes an implicit part of NumPy's public API. For example, -refactoring ``stack`` to use ``np.block()`` instead of -``np.concatenate()`` internally would now become a breaking change. 2. -Array libraries may prefer to implement high level functions differently -than NumPy. For example, a library might prefer to implement a -fundamental operations like ``mean()`` directly rather than relying on -``sum()`` followed by division. More generally, it's not clear yet what -exactly qualifies as core functionality, and figuring this out could be -a large project. 3. We don't yet have an overloading system for -attributes and methods on array objects, e.g., for accessing ``.dtype`` -and ``.shape``. This should be the subject of a future NEP, but until -then we should be reluctant to rely on these properties. +However, this also comes with several downsides: + +1. The details of how NumPy implements a high-level function in terms of + overloaded functions now becomes an implicit part of NumPy's public API. For + example, refactoring ``stack`` to use ``np.block()`` instead of + ``np.concatenate()`` internally would now become a breaking change. +2. Array libraries may prefer to implement high level functions differently than + NumPy. For example, a library might prefer to implement a fundamental + operations like ``mean()`` directly rather than relying on ``sum()`` followed + by division. More generally, it's not clear yet what exactly qualifies as + core functionality, and figuring this out could be a large project. +3. We don't yet have an overloading system for attributes and methods on array + objects, e.g., for accessing ``.dtype`` and ``.shape``. This should be the + subject of a future NEP, but until then we should be reluctant to rely on + these properties. Given these concerns, we encourage relying on this approach only in limited cases. @@ -438,25 +497,50 @@ make it impossible to implement this generic fallback behavior for ``__array_function__``. We could resolve this issue by change the handling of return values in -``__array_function__`` in either of two possible ways: 1. Change the -meaning of all arguments returning ``NotImplemented`` to indicate that -all arguments should be coerced to NumPy arrays instead. However, many -array libraries (e.g., scipy.sparse) really don't want implicit -conversions to NumPy arrays, and often avoid implementing ``__array__`` -for exactly this reason. Implicit conversions can result in silent bugs -and performance degradation. 2. Use another sentinel value of some sort -to indicate that a class implementing part of the higher level array API -is coercible as a fallback, e.g., a return value of -``np.NotImplementedButCoercible`` from ``__array_function__``. - -If we take this second approach, we would need to define additional -rules for how coercible array arguments are coerced, e.g., - Would we -try for ``__array_function__`` overloads again after coercing coercible -arguments? - If so, would we coerce coercible arguments one-at-a-time, -or all-at-once? - -These are slightly tricky design questions, so for now we propose to -defer this issue. We can always implement +``__array_function__`` in either of two possible ways: + +1. Change the meaning of all arguments returning ``NotImplemented`` to indicate + that all arguments should be coerced to NumPy arrays instead. However, many + array libraries (e.g., scipy.sparse) really don't want implicit conversions + to NumPy arrays, and often avoid implementing ``__array__`` for exactly this + reason. Implicit conversions can result in silent bugs and performance + degradation. +2. Use another sentinel value of some sort, e.g., + ``np.NotImplementedButCoercible``, to indicate that a class implementing part + of NumPy's higher level array API is coercible as a fallback. This is a more + appealing option. + +If we take this second approach, we would need to define additional rules for +how coercible array arguments are coerced. For sanity, a return value of +``np.NotImplementedButCoercible`` from ``__array_function__`` should be handled +equivalently to the corresponding argument not defining an +``__array_function__`` method at all. If all arguments to a NumPy function +either have no ``__array_function__`` method defined or return +``NotImplementedButCoercible`` from ``__array_function__``, we would fall back +to NumPy's current behavior: all arguments would be coerced to NumPy arrays. + +This approach also suggests an intriguing possibility: default implementations +of ``__array_function__`` and ``__array_ufunc__`` on ``numpy.ndarray`` could +change to simply return ``NotImplementedButCoercible``, i.e., + +.. code:: python + + class ndarray: + def __array_ufunc__(self, *args, **kwargs): + return NotImplementedButCoercible + def __array_function__(self, *args, **kwargs): + return NotImplementedButCoercible + +This logic is not only much simpler than the existing implementation of +``numpy.ndarray.__array_function__``, but it also means we could remove all +special cases for ``numpy.ndarray`` in the NumPy's dispatch logic. It would +break `some existing implementations <https://mail.python.org/pipermail/numpy-discussion/2018-June/078175.html>`_ +of ``__array_ufunc__`` on ndarray subclasses (e.g., in astropy), but this +would be acceptable given the still experimental status of``__array_ufunc__``. + +It is not yet clear to us if we need ``NotImplementedButCoercible`` or should +change ``ndarray.__array_function__`` as shown above, so for +now we propose to defer this issue. We can always implement ``np.NotImplementedButCoercible`` at some later time if it proves critical to the numpy community in the future. Importantly, we don't think this will stop critical libraries that desire to implement most of @@ -512,32 +596,21 @@ Customizing the detailed behavior of array libraries will require using library specific functions, which could be limiting in the case of libraries that consume the NumPy API such as xarray. - Discussion ---------- Various alternatives to this proposal were discussed in a few Github issues: -1. `pydata/sparse #1 <https://github.com/pydata/sparse/issues/1>`_ -2. `numpy/numpy #11129 <https://github.com/numpy/numpy/issues/11129>`_ +1. `pydata/sparse #1 <https://github.com/pydata/sparse/issues/1>`_ +2. `numpy/numpy #11129 <https://github.com/numpy/numpy/issues/11129>`_ Additionally it was the subject of `a blogpost -<http://matthewrocklin.com/blog/work/2018/05/27/beyond-numpy>`_ Following this +<http://matthewrocklin.com/blog/work/2018/05/27/beyond-numpy>`_. Following this it was discussed at a `NumPy developer sprint <https://scisprints.github.io/#may-numpy-developer-sprint>`_ at the `UC Berkeley Institute for Data Science (BIDS) <https://bids.berkeley.edu/>`_. - -References and Footnotes ------------------------- - -.. [1] Each NEP must either be explicitly labeled as placed in the public domain (see - this NEP as an example) or licensed under the `Open Publication License`_. - -.. _Open Publication License: http://www.opencontent.org/openpub/ - - Copyright --------- -This document has been placed in the public domain. [1]_ +This document has been placed in the public domain. |