On Sun, Jan 21, 2018 at 9:48 PM, Allan Haldane <allanhaldane@gmail.com> wrote:

Hello all,

We are making a decision (again) about what to do about the
behavior of multiple-field indexing of structured arrays: Should
it return a view or a copy, and on what release schedule?

As a reminder, this refers to operations like (1.13 behavior):

    >>> a = np.zeros(3, dtype=[('a', 'i4'), ('b', 'i4'), ('c', 'f4')])
    >>> a[['a', 'c']]
    array([(0, 0.), (0, 0.), (0, 0.)],
          dtype=[('a', '<i4'), ('c', '<f4')]

In numpy 1.14.0 we made this return a view instead of a copy, but
downstream test failures suggest we reconsider. In our current
implementation for 1.14.1, we have reverted this change, but
still plan to go through with it in 1.15.

See here for our discussion the problem and solutions:
https://github.com/numpy/numpy/pull/10411

The two main options we have discussed are either to try to make
the change in 1.15, or never make the change at all and always
return a copy.

Here are some pros and cons:

Pros (change to view in 1.15)
=============================

 * Views are useful and convenient. Other forms of indexing also
   often return views so this is more consistent.
 * This change has been planned since numpy 1.7 in 2009,
   and there have been visible FutureWarnings about it since
   then. Anyone whose code will break should have seen the
   warnings. It has been extensively warned about in recent
   release notes.
 * Past discussions have supported the change. See my comment in
   the PR with many links to them and to other history.
 * Users have requested the change on the list.
 * Possibly a majority of the reported code failures were not
   actually caused by the change, but by another bug (#8100)
   involving np.load/np.save which this change exposed. If we
   push it off to 1.15, we will have time to fix this other bug.
   (There were no FutureWarnings for this breakage, of course).
 * The code that really will break is of the form
         a[['a', 'c']].view('i8')
   because the returned itemsize is different. This has
   raised FutureWarnings since numpy 1.7, and no users reported
   failures due to this change. In the PR we still try to
   mitigate this breakage by introducing a new method
   `pack_fields`, which converts the result into the 1.13 form,
   so that
         np.pack_fields(a[['a', 'c']]).view('i8')
   will work.


Cons (keep returning a copy)
============================

 * The extra convenience is not really that much, and fancy
   indexing also returns a copy instead of a view, so there is
   a precedent there.
 * We want to minimize compatibility breaks with old behavior.
   We've had a fair amount of discussion and complaints about
   how we break things in general.
 * We have lived with a "copy" for 8 years now. At some point the
   behavior gets set in stone for compatibility reasons.
 * Users have written to the list and github about their code
   breaking in 1.14.0. As far as I am aware, they all refer
   to the #8100 problem.
 * If a new function `pack_fields` is needed to guard against
   mishaps with the view behavior, that seems like a sign that
   keeping the copy behavior is the best option from an API
   perspective.

My initial vote is go with the change in 1.15: The "view" code
that will ultimately break (not the code related to #8100) has
been sending FutureWarnings for many years, and I am not aware of
any user complaints involving it: All the complaints so far
would be fixed with #8100 in 1.15.

(Note based on a linked mailing list thread, 2012 might be the last time I looked more closely at structured dtypes. 

So some of what I understand might be outdated.)

views on structured dtypes are very important, but viewing them as standard arrays with standard dtypes is the main part that I had used.

Essentially structured dtypes are useless for any computation, e.g. just some simple reduce operation. To work with them we need a standard view.

I think the usecase that fails in statsmodels (except there is no test failure anymore because we switched to using pandas in the unit test)

        cls.confint_res = cls.results[['acvar_lb','acvar_ub']].view((float,

>                                                                           2))

E       ValueError: Changing the dtype to a subarray type is only supported if the total itemsize is unchanged

This is similar to the above example

a[['a', 'c']].view('i8')

but it doesn't try to combine fields. 

In  many examples where I used structured dtypes a long time ago, switched between consistent views as either a standard array of subsets or as .structured dtypes.

For this usecase it wouldn't matter whether a[['a', 'c']] returns a view or copy, as long as we can get the second view that is consistent with the selected part of the memory. This would also be independent of whether numpy pads internally and adjusts the strides if possible or not.

>>> np.__version__

'1.11.2'

>>> a = np.ones(5, dtype=[('a', 'i8'), ('b', 'f8'), ('c', 'f8')])

>>> a

array([(1, 1.0, 1.0), (1, 1.0, 1.0), (1, 1.0, 1.0), (1, 1.0, 1.0),

       (1, 1.0, 1.0)], 

      dtype=[('a', '<i8'), ('b', '<f8'), ('c', '<f8')])

>>> a.mean(0)

Traceback (most recent call last):

  File "<pyshell#15>", line 1, in <module>

    a.mean(0)

  File "C:\...\python-3.4.4.amd64\lib\site-packages\numpy\core\_methods.py", line 65, in _mean

    ret = umr_sum(arr, axis, dtype, out, keepdims)

TypeError: cannot perform reduce with flexible type

>>> a[['b', 'c']].mean(0)

Traceback (most recent call last):

  File "<pyshell#16>", line 1, in <module>

    a[['b', 'c']].mean(0)

  File "C:\...\python-3.4.4.amd64\lib\site-packages\numpy\core\_methods.py", line 65, in _mean

    ret = umr_sum(arr, axis, dtype, out, keepdims)

TypeError: cannot perform reduce with flexible type

>>> a[['b', 'c']].view(('f8', 2)).mean(0)

array([ 1.,  1.])

>>> a[['b', 'c']].view(('f8', 2)).dtype

dtype('float64')

Aside The plan is that statsmodels will drop all usage and support for rec_arays/structured dtypes 

in the following release (0.10).
Then structured dtypes are free (from our perspective) to provide low level struct support
instead of pretending to be dataframe_like.

Josef

 

Feel free to also discuss the related proposed change, to make
np.diag return a view instead of a copy. That change has
not been implemented yet, only proposed. 

Cheers,
Allan
_______________________________________________
NumPy-Discussion mailing list
NumPy-Discussion@python.org
https://mail.python.org/mailman/listinfo/numpy-discussion