I think you are saying that this current behavior of `choose` should be considered a bug. I hope not, because as illustrated, it is useful. How would you propose to efficiently do the same substitutions? On 4/17/2021 4:27 PM, Kevin Sheppard wrote:

2. I would describe this a a bug. I think sequences are converted to arrays and in this case the conversion is not returning a 2 element object array but expanding and then concatenation.

On Sat, Apr 17, 2021, 18:56 Alan G. Isaac mailto:alan.isaac@gmail.com> wrote:

2. Separately, mypy is unhappy with my 2nd argument to `choose`: Argument 2 to "choose" has incompatible type "Tuple[int, Sequence[float]]"; expected "Union[Union[int, float, complex, str, bytes, generic], Sequence[Union[int, float, complex, str, bytes, generic]], Sequence[Sequence[Any]],_SupportsArray]" However, `choose` is happy to have e.g. `choices=(0,seq)` (and I hope it will remain so!).

E.g., a = a = (0,1,1,0,0,0,1,1)  #binary array np.choose(a, (0,range(8))     #array([0, 1, 2, 0, 0, 0, 6, 7])

Oh. I answered thinking about choice and not choose. Please ignore both parts. On Sun, Apr 18, 2021, 17:56 Robert Kern wrote:

On Sat, Apr 17, 2021 at 4:28 PM Kevin Sheppard wrote:

...

1. I suppose it only uses the (Native int or int64) dtype since each one would need a code path to run quickly.

2. I would describe this a a bug. I think sequences are converted to arrays and in this case the conversion is not returning a 2 element object array but expanding and then concatenation.

No, it's broadcasting the two to a common shape, as documented. https://numpy.org/doc/stable/reference/generated/numpy.choose.html

...

E.g.,

...

a = a = (0,1,1,0,0,0,1,1) #binary array np.choose(a, (0,range(8)) #array([0, 1, 2, 0, 0, 0, 6, 7])

This is equivalent to `np.choose(a, (np.zeros(8, dtype=int), range(8)))`

-- Robert Kern _______________________________________________ NumPy-Discussion mailing list NumPy-Discussion@python.org https://mail.python.org/mailman/listinfo/numpy-discussion

On Mon, Apr 19, 2021 at 12:24 PM bas van beek wrote:

After a taking a closer look at the `np.choose` docs I think we should change `*choices: npt.ArrayLike` into `choices: Sequence[npt.ArrayLike]`.*

*This would resolve the issue, but it’d also mean that directly passing an array will be prohibited (as far as type checkers are concerned).*

*The docs do mention that the outermost container should be a list or tuple anyway, so I’m not convinced that this new typing restriction would be a huge loss.*

Why not allow both? -- Robert Kern